WO2009042115A2 - Procédés de mise au point d'arn en épingle à cheveux courts (arnsh) pour le silençage de gènes - Google Patents

Procédés de mise au point d'arn en épingle à cheveux courts (arnsh) pour le silençage de gènes Download PDF

Info

Publication number
WO2009042115A2
WO2009042115A2 PCT/US2008/011010 US2008011010W WO2009042115A2 WO 2009042115 A2 WO2009042115 A2 WO 2009042115A2 US 2008011010 W US2008011010 W US 2008011010W WO 2009042115 A2 WO2009042115 A2 WO 2009042115A2
Authority
WO
WIPO (PCT)
Prior art keywords
sequence
sirna
shrna
motifs
sirnas
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/US2008/011010
Other languages
English (en)
Other versions
WO2009042115A3 (fr
Inventor
Julja Burchard
Michele A. Cleary
Wei Ge
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Rosetta Inpharmatics LLC
Original Assignee
Rosetta Inpharmatics LLC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Rosetta Inpharmatics LLC filed Critical Rosetta Inpharmatics LLC
Publication of WO2009042115A2 publication Critical patent/WO2009042115A2/fr
Publication of WO2009042115A3 publication Critical patent/WO2009042115A3/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • C—CHEMISTRY; METALLURGY
    • C12—BIOCHEMISTRY; BEER; SPIRITS; WINE; VINEGAR; MICROBIOLOGY; ENZYMOLOGY; MUTATION OR GENETIC ENGINEERING
    • C12N—MICROORGANISMS OR ENZYMES; COMPOSITIONS THEREOF; PROPAGATING, PRESERVING, OR MAINTAINING MICROORGANISMS; MUTATION OR GENETIC ENGINEERING; CULTURE MEDIA
    • C12N15/00—Mutation or genetic engineering; DNA or RNA concerning genetic engineering, vectors, e.g. plasmids, or their isolation, preparation or purification; Use of hosts therefor
    • C12N15/09—Recombinant DNA-technology
    • C12N15/11—DNA or RNA fragments; Modified forms thereof; Non-coding nucleic acids having a biological activity
    • C12N15/111—General methods applicable to biologically active non-coding nucleic acids
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B30/00—ICT specially adapted for sequence analysis involving nucleotides or amino acids
    • G16B30/10—Sequence alignment; Homology search
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/20—Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/30—Detection of binding sites or motifs
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/50—Mutagenesis
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B30/00—ICT specially adapted for sequence analysis involving nucleotides or amino acids
    • C—CHEMISTRY; METALLURGY
    • C12—BIOCHEMISTRY; BEER; SPIRITS; WINE; VINEGAR; MICROBIOLOGY; ENZYMOLOGY; MUTATION OR GENETIC ENGINEERING
    • C12N—MICROORGANISMS OR ENZYMES; COMPOSITIONS THEREOF; PROPAGATING, PRESERVING, OR MAINTAINING MICROORGANISMS; MUTATION OR GENETIC ENGINEERING; CULTURE MEDIA
    • C12N2310/00—Structure or type of the nucleic acid
    • C12N2310/10—Type of nucleic acid
    • C12N2310/14—Type of nucleic acid interfering nucleic acids [NA]
    • C—CHEMISTRY; METALLURGY
    • C12—BIOCHEMISTRY; BEER; SPIRITS; WINE; VINEGAR; MICROBIOLOGY; ENZYMOLOGY; MUTATION OR GENETIC ENGINEERING
    • C12N—MICROORGANISMS OR ENZYMES; COMPOSITIONS THEREOF; PROPAGATING, PRESERVING, OR MAINTAINING MICROORGANISMS; MUTATION OR GENETIC ENGINEERING; CULTURE MEDIA
    • C12N2310/00—Structure or type of the nucleic acid
    • C12N2310/50—Physical structure
    • C12N2310/53—Physical structure partially self-complementary or closed
    • C—CHEMISTRY; METALLURGY
    • C12—BIOCHEMISTRY; BEER; SPIRITS; WINE; VINEGAR; MICROBIOLOGY; ENZYMOLOGY; MUTATION OR GENETIC ENGINEERING
    • C12N—MICROORGANISMS OR ENZYMES; COMPOSITIONS THEREOF; PROPAGATING, PRESERVING, OR MAINTAINING MICROORGANISMS; MUTATION OR GENETIC ENGINEERING; CULTURE MEDIA
    • C12N2320/00—Applications; Uses
    • C12N2320/10—Applications; Uses in screening processes
    • C12N2320/11—Applications; Uses in screening processes for the determination of target sites, i.e. of active nucleic acids
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations

Definitions

  • the invention relates to methods for designing shRNAs with higher overall silencing efficacy.
  • the invention also relates to methods for designing DNA constructs that can express shRNAs with high transcription efficiencies.
  • RNA interference suppresses gene expression in cells from various organisms including mammalian cells, and has generated much excitement in the scientific community (Couzin, 2002, Science 298:2296-2297; McManus et al. , 2002, Nat. Rev. Genet. 3:737-747; Hannon, G. J., 2002, Nature 418:244-251; Paddison et al., 2002, Cancer Cell 2:17-23).
  • RNAi is conserved throughout evolution, from C. elegans to humans, and is believed to function in protecting cells from invasion by RNA viruses.
  • dsRNA double stranded RNA
  • Dicer RNaselll-type enzyme
  • the Dicer enzyme "dices" the RNA into short duplexes of 21 to 23 nucleotides (nt), termed short-interfering RNAs (siRNAs), which are composed of 19 to 21 nt of perfectly paired ribonucleotides with two unpaired nucleotides on the 3' end of each strand.
  • siRNAs short-interfering RNAs
  • RISC RNA-induced silencing complex
  • nucleases present in the RISC complex cleave the mRNA transcripts, thereby abolishing expression of the gene product. In the case of viral infection, this mechanism would result in destruction of viral transcripts, thus preventing viral synthesis. Since the siRNAs are double-stranded, either strand has the potential to associate with RISC and direct silencing of transcripts with sequence similarity.
  • miRNAs are regulatory RNAs expressed from the genome, and are processed from precursor stem-loop structures to produce single-stranded nucleic acids that bind to sequences in the 3' UTR of the target mRNA (Lee et al, 1993, Cell 75:843-854; Reinhart et al, 2000, Nature 403:901-906; Lee et al, 2001, Science 294:862-864; and Lau et al, 2001, Science 294:858-862; Hutvagner et al, 2001, Science 293:834-838).
  • miRNAs bind to transcript sequences with only partial complementarity (Zeng et al , 2002, Molec. Cell 9:1327-1333) and repress translation without affecting steady-state RNA levels (Lee et al, 1993, Cell 75:843-854; and Wightman et al, 1993, Cell 75:855-862). Both miRNAs and siRNAs are produced by Dicer and associate with components of the RISC complex (Hutvagner et al, 2001, Science 293:834-838; Grishok et al, 2001, Cell 106: 23-34; Ketting et al, 2001, Genes Dev. 15:2654-2659; Williams et al, 2002, Proc. Natl Acad. Sci.
  • siRNAs for gene silencing in vivo enables possible selection and development of siRNAs for therapeutic use.
  • a recent report highlights the potential therapeutic application of siRNAs. Fas-mediated apoptosis is implicated in a broad spectrum of liver diseases, where lives could be saved by inhibiting apoptotic death of hepatocytes.
  • the Fas gene was silenced in mouse hepatocytes at the mRNA and protein levels, thereby preventing apoptosis and protecting the mice from hepatitis-induced liver damage (Song et ah, 2003, Nat. Medicine 9:347-351).
  • silencing Fas expression holds therapeutic promise to prevent liver injury by protecting hepatocytes from cytotoxicity.
  • mice were injected intraperitoneally with siRNAs targeting TNF-a. Lipopolysaccharide-induced TNF-a gene expression was inhibited, and these mice were protected from sepsis.
  • United States Publication No. 6,506,559 discloses an RNA interference process for inhibiting expression of a target gene in a living cell.
  • the patent discloses introducing partially or fully doubled-stranded RNAs having a sequence in the duplex region that is identical to a sequence in the target gene into the living cell or into the extracellular environment. RNA sequences with insertions, deletions, and single point mutations relative to the target sequence are also disclosed as effective for expression inhibition.
  • United States Patent Publication No. US 2002/0086356 discloses RNA interference in a Drosophila in vitro system using RNA segments of 21 -23 nt in length.
  • the patent publication teaches that when these 21-23 nt fragments are purified and added back to Drosophila extracts, they mediate sequence-specific RNA interference in the absence of long double-stranded RNAs (dsRNAs).
  • dsRNAs long double-stranded RNAs
  • the patent publication also teaches that chemically synthesized oligonucleotides (oligos) of the same or similar nature can be used to target specific mRNAs for degradation in mammalian cells.
  • dsRNAs double-stranded RNAs
  • siRNAs duplexes can be generated by an RNase Ill-like processing reaction from long dsRNAs or by chemically synthesized siRNA duplexes with overhanging 3' ends mediating efficient target RNA cleavage in the lysate where the cleavage site is located near the center of the region spanned by the guiding siRNA.
  • the PCT publication also provides evidence that the direction of dsRNA processing determines whether sense or antisense-identical target RNA can be cleaved by the produced siRNA complex.
  • Systematic analyses of the effects of length, secondary structure, sugar backbone and sequence specificity of siRNAs on RNA interference have been disclosed to aid siRNA design.
  • silencing efficacy has been shown to correlate with the GC content of the 5' and 3' regions of the 19 base pair target sequence. It was found that siRNAs targeting sequences with a GC rich 5' and GC poor 3' perform the best. More detailed discussion may be found in Elbashir et al, 2001, EMBO J. 20:6877-6888 and Aza-Blanc et al, 2003, MoI. Cell 12:627-637; each of which is hereby incorporated by reference herein in its entirety.
  • siRNA design algorithms are disclosed in PCT publications WO 2005/018534 A2 and WO 2005/042708 A2; each of which is hereby incorporated by reference herein in its entirety.
  • WO 2005/018534 A2 discloses methods and compositions for gene silencing using siRNA having partial sequence homology to its target gene.
  • the application provides methods for identifying common and / or differential responses to different siRNAs targeting a gene.
  • the application also provides methods for evaluating the relative activity of the two strands of an siRNA.
  • the application further provides methods of using siRNAs as therapeutics for treatment of diseases.
  • WO 2005/042708 A2 provides a method for identifying siRNA target motifs in a transcript using a position-specific score matrix approach. It also provides a method for identifying off-target genes of an siRNA using a position-specific score matrix approach.
  • the application further provides a method for designing siRNAs with improved silencing efficacy and specificity as well as a library of exemplary siRNAs.
  • RNA molecules are limited. For example, in cell lines that are difficult to transfect, siRNA duplexes may not work at all. In addition, siRNAs can be expensive because oligos must be purchased or synthesized for every experiment. It is also difficult to trace or quantify oligo delivery, making normalization difficult.
  • shRNA DNA expression vector based short hairpin RNA
  • An shRNA molecule contains a sense strand and an antisense strand, and a loop sequence between the sense and antisense strands. Due to the sequence complementarity of the sense and antisense strands, such RNA molecules form hairpin- shaped double-stranded RNAs (dsRNAs).
  • dsRNAs hairpin- shaped double-stranded RNAs
  • sequence encoding an shRNA is cloned into a vector and the vector is introduced into a cell and transcribed by the cell's transcription machinery (Chen et al, 2003, Biochem Biophys Res Commun 311 :398-404).
  • the shRNAs can then be transcribed, for example, by RNA polymerase III (Pol III) in response to a Pol Ill-type promoter in the vector (Yuan et al, 2006, MoI Biol Rep 33:33- 41 and Scherer et al, 2004, MoI Ther 10:597-603).
  • the expressed shRNAs are then exported into the cytoplasm where they are processed by proteins such as Dicer into siRNAs, which then trigger RNAi (Amarzguioui et al. , 2005, FEBS Letter 579:5974- 5981). It has been reported that purines are required at the 5' end of a newly initiated RNA for optimal RNA polymerase III transcription.
  • shRNAs expressed from a DNA construct by RNA polymerase III (Pol III) transcription the transcription efficiency depends on whether the transcription start site contains purines or pyrimidines (see, e.g., Zecherle et al, 1996, Molecular and Cellular Biology, 16:5801-5810).
  • the vector based expression of shRNA core sequences is advantageous over the oligo-based siRNAs for several reasons. It is more versatile than oligo- based siRNAs.
  • the shRNAs core sequences can be expressed stably in cells, allowing long-term gene silencing in cells both in vitro and in vivo, e.g., in animals (see, McCaffrey et al, 2002, Nature 418:38-39; Xia et al, 2002, Nat. Biotech. 20:1006-1010; Lewis et al, 2002, Nat. Genetics 32:107- 108; Rubinson et al, 2003, Nat. Genetics 33:401-406; and Tiscornia et al, 2003, Proc. Natl. Acad. Sci. USA 100:1844-1848).
  • shRNA core sequences allow both long-term and transient gene expression for better controlled gene silencing than the oligo-based siRNA technology. Because DNA handling and delivery methods are well-established, long term shRNA-based RNAi experiments are more cost- effective. In addition, transfection efficiency can be normalized to accurately quantify reduction of gene expression (gene knockdown). Sophisticated DNA handling tools allow fluorescently tagged fusion proteins to be tagged to the corresponding siRNA duplex to easily trace sequence delivery. It is also possible to use flow cytometry to enrich transfected populations.
  • the silencing efficacies of shRNA core sequences are not as potent as siRNA duplexes harboring the same nucleotide sequence obtained by oligo synthesis.
  • Multiple factors may account for the attenuated silencing efficacies of shRNAs when compared with their corresponding siRNAs.
  • the silencing efficacy of an shRNA core sequence may be compromised if the DNA construct encoding the shRNA core sequence is not properly or efficiently transcribed. What is needed in the art are systems and methods for improving the silencing efficacies of shRNAs.
  • the present invention provides methods for classifying siRNA functional sequence motifs for use in an shRNA or shRNA core sequence designed to produce an siRNA for silencing a target gene in a cell.
  • the shRNA core sequence comprises an antisense strand sequence and a sense strand sequence.
  • the invention provides a method for selecting (a) selecting a first plurality of siRNA functional sequence motifs, wherein each siRNA functional sequence motif of said first plurality of siRNA functional sequence motifs has a predicted silencing efficacy against a targeted sequence motif within a transcript of a target gene and said predicted silencing efficacy is equal or greater than a chosen threshold of silencing efficacy; and (b) classifying each siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs as belonging to a first class of siRNA functional sequence motifs when said siRNA functional sequence motif has a 3' terminal nucleotide that is a cytosine, or as belonging to a second class of siRNA functional sequence motifs when said siRNA functional sequence motif does not have a 3 ' terminal nucleotide that is a cytosine.
  • the chosen threshold is 50%, 75%, 80% or 90% at an siRNA concentration of 100 nM in culture medium in which a cell is expressing said target gene is incuba
  • the method further comprises (a ⁇ ) determining, prior to said selecting step (a), said predicted silencing efficacy of an siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs based on a PSSM matrix. In other embodiments, the method further comprises (a ⁇ ) determining, prior to said selecting step (a), said predicted silencing efficacy of an siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs based on a predicted binding affinity between said siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs and a susceptible sequence motif of said target gene.
  • the method further comprises (a ⁇ ) determining, prior to said selecting step (a), said predicted silencing efficacy of an siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs based on an observed silencing efficacy of said siRNA functional sequence motif.
  • an siRNA functional sequence motif of said first plurality of siRNA functional sequence motifs has at least 19 nucleotides, at least 21 nucleotides, or least 27 nucleotides.
  • the method further comprises (iiil) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; said sense strand sequence and /or said antisense strand sequence.
  • the present invention also provides methods of designing an shRNA or shRNA core sequence or sense strand sequence thereof or antisense strand sequence thereof, wherein said shRNA core sequence is designed to produce an siRNA for silencing a target gene in a cell of an organism.
  • the shRNA core sequence comprises an antisense strand sequence and a sense strand sequence.
  • the method comprises: (i) carrying out any method comprising any one of the selecting (a), classifying (b) or determining (a ⁇ ) steps; (ii) selecting an siRNA functional sequence strand sequence in an shRNA core sequence; and (iii) designing a sense strand sequence based on said antisense strand sequence in said shRNA core sequence, wherein said sense strand sequence is an inverted repeat to said antisense strand sequence.
  • the method further comprises (iv) designing a nucleotide sequence (hereinafter "loop sequence") in between said antisense strand sequence and said sense strand sequence, thereby completing said shRNA core sequence.
  • the loop sequence consists of at least 3 nucleotides, at least 7 nucleotides, at least 9 nucleotides, at least 15 nucleotides or at least 25 nucleotides.
  • the method further comprises (v) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; one or more of said loop sequence, said sense strand sequence, said antisense strand sequence, and said shRNA core sequence.
  • the present invention also provides methods of designing an shRNA or shRNA core sequence or sense strand sequence thereof or antisense strand sequence thereof, wherein said shRNA core sequence is designed to produce an siRNA for silencing a target gene in a cell.
  • the shRNA core sequence comprises an antisense strand sequence and a sense strand sequence.
  • the method comprises: (i) carrying out any method comprising any one of the selecting (a), classifying (b) or determining (a ⁇ ) steps; (ii) selecting an siRNA functional sequence motif from said second class of siRNA functional sequence motifs; (iii) replacing the nucleotide at the 3' terminus of said siRNA functional motif with a cytosine, thereby creating a new siRNA functional sequence motif to be an antisense strand sequence in an shRNA core sequence; and (iv) designing a sense strand sequence based on said antisense strand sequence in said shRNA core sequence, wherein said sense strand sequence is an inverted repeat to said antisense strand sequence.
  • the method further comprises (v) designing a nucleotide sequence (hereinafter “loop sequence”) in between said antisense strand sequence and said sense strand sequence, thereby completing said shRNA core sequence.
  • the loop sequence consists of at least 3 nucleotides, at least 7 nucleotides, at least 9 nucleotides, at least 15 nucleotides or at least 25 nucleotides.
  • the method further comprises (el) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; said sense strand sequence and / or said antisense strand sequence. In some embodiments, the method further comprises (fl) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; one or more of said loop sequence, said sense strand sequence, said antisense strand sequence, and said shRNA core sequence.
  • the present invention also provides methods of designing an shRNA or shRNA core sequence or sense strand sequence thereof or antisense strand sequence thereof, wherein said shRNA core sequence is designed to produce an siRNA for silencing a target gene in a cell.
  • the shRNA core sequence comprises an antisense strand sequence and a sense strand sequence.
  • the method comprises (a) selecting a first plurality of siRNA functional sequence motifs, wherein each siRNA functional sequence motif of said first plurality of siRNA functional sequence motifs has a predicted silencing efficacy against a targeted sequence motif within a transcript of a target gene and said predicted silencing efficacy is equal or greater than a chosen threshold of silencing efficacy; (b) selecting an siRNA functional sequence motif from said first plurality of siRNA functional sequence motifs; (c) adding a cytosine to the 3' terminus of said siRNA functional sequence motif, thereby creating a new siRNA functional sequence motif to be an antisense strand sequence in an shRNA core sequence; and (d) designing a sense strand sequence based on said antisense strand sequence in said shRNA core sequence, wherein said sense strand sequence is an inverted repeat to said antisense strand sequence.
  • the method further comprise (e) designing a nucleotide sequence (hereinafter "loop sequence") in between said antisense strand sequence and said sense strand sequence, thereby completing said shRNA core sequence.
  • the loop sequence consists of at least 3 nucleotides, at least 7 nucleotides, at least 9 nucleotides, at least 15 nucleotides or at least 25 nucleotides.
  • the method further comprises (dl) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; said sense strand sequence and / or said antisense strand sequence.
  • the method further comprises (el) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; one or more of said loop sequence, said sense strand sequence, said antisense strand sequence, and said shRNA core sequence.
  • the method further comprises (dl) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; said sense strand sequence and / or said antisense strand sequence. In some embodiments, the method further comprises (el) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; one or more of said loop sequence, said sense strand sequence, said antisense strand sequence, and said shRNA core sequence.
  • the present invention provides methods of designing an shRNA or an shRNA core sequence or sense strand sequence thereof or antisense strand sequence thereof or designing siRNA functional sequence motifs for use in an shRNA core sequence where the shRNA core sequence is designed to produce an siRNA for silencing a target gene in a cell of an organism.
  • the shRNA core sequence comprises an antisense strand sequence and a sense strand sequence.
  • the method comprises (a) selecting a first plurality of siRNA functional sequence motifs, wherein each siRNA functional sequence motif of said first plurality of siRNA functional sequence motifs has a predicted silencing efficacy against a targeted sequence motif within a transcript of a target gene and said predicted silencing efficacy is equal or greater than a chosen threshold of silencing efficacy; and (b) deleting from said first plurality of siRNA functional sequence motifs each siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs that does not have a 3 ' terminal nucleotide that is a cytosine, thereby forming a first class of siRNA functional sequence motifs with a 3' terminal cytosine.
  • the chosen threshold is 50%, 75%, 80% or 90% at an siRNA concentration of 100 nM in culture medium in which a cell is expressing said target gene is incubated.
  • the method further comprises (a ⁇ ) determining, prior to said selecting step (a), said predicted silencing efficacy of an siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs based on a PSSM matrix.
  • the method further comprises (a ⁇ ) determining, prior to said selecting step (a), said predicted silencing efficacy of an siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs based on a predicted binding affinity between said siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs and a susceptible sequence motif of said target gene.
  • the method further comprises (a ⁇ ) determining, prior to said selecting step (a), said predicted silencing efficacy of an siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs based on an observed silencing efficacy of said siRNA functional sequence motif.
  • the loop sequence consists of at least 3 nucleotides, at least 7 nucleotides, at least 9 nucleotides, at least 15 nucleotides or at least 25 nucleotides.
  • the method further comprises (c) selecting an siRNA functional sequence motif from said first class of siRNA functional sequence motifs to be an antisense strand sequence in an shRNA core sequence; and (d) designing a sense strand sequence based on said antisense strand sequence in said shRNA core sequence, wherein said sense strand sequence is an inverted repeat to said antisense strand sequence.
  • the method further comprises (e) designing a nucleotide sequence
  • the method further comprises (bl) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; said first class of siRNA functional sequence motifs or one or more siRNA functional sequence motifs in said first class of siRNA functional sequence motifs.
  • the method further comprises (dl) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; said sense strand sequence and / or said antisense strand sequence.
  • the present invention also provides methods of designing an shRNA or shRNA core sequence or sense strand sequence thereof or antisense strand sequence thereof or designing siRNA functional sequence motifs for use in an shRNA core sequence, where said shRNA core sequence is designed to produce an siRNA for silencing a target gene in a cell of an organism.
  • the shRNA core sequence comprises an antisense strand sequence and a sense strand sequence.
  • the method comprises (a) selecting a first plurality of siRNA functional sequence motifs, wherein each siRNA functional sequence motif of said first plurality of siRNA functional sequence motifs has a predicted silencing efficacy against a targeted sequence motif within a transcript of a target gene and said predicted silencing efficacy is equal or greater than a chosen threshold of silencing efficacy; and (b) deleting from said first plurality of siRNA functional sequence motifs each siRNA functional sequence motif in said first plurality of siRNA functional sequence motifs that has a 3 ' terminal nucleotide that is a cytosine, thereby forming a second class of siRNA functional sequence motifs that do not have a 3' terminal cytosine.
  • the chosen threshold is 50%, 75%, 80% or 90% at an siRNA concentration of 100 nM in culture medium in which a cell is expressing said target gene is incubated.
  • the method further comprises (c) selecting an siRNA functional sequence motif from said second class of siRNA functional sequence motifs; (d) replacing the nucleotide at the 3' terminus of said siRNA functional motif with a cytosine, thereby creating a new siRNA functional sequence motif to be an antisense strand sequence in an shRNA core sequence; and (e) designing a sense strand sequence based on said antisense strand sequence in said shRNA core sequence, wherein said sense strand sequence is an inverted repeat to said antisense strand sequence.
  • the method further comprises (f) designing a nucleotide sequence (hereinafter "loop sequence") in between said antisense strand sequence and said sense strand sequence, thereby completing said shRNA core sequence.
  • the loop sequence consists of at least 3 nucleotides, at least 7 nucleotides, at least 9 nucleotides, at least 15 nucleotides or at least 25 nucleotides.
  • the method further comprises (el) outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; one or more of said loop sequence, said sense strand sequence, said antisense strand sequence, and said shRNA core sequence.
  • the method further comprises adding a flanking region at the 5' terminus of the shRNA core sequence; adding a flanking region at the 3' terminus of the shRNA core sequence; adding a first flanking region at the 3' terminus of said shRNA core sequence, and a second flanking region at the 5' terminus of said shRNA core sequence; or adding a promoter sequence 5' to said shRNA core sequence.
  • the promoter sequence is selected from the group consisting of a human, bovine or mouse Hl promoter, a human, bovine or mouse U6 promoter, and a human, bovine or mouse U2 promoter.
  • the method further comprises outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; said flanking region(s) at the 5' terminus and / or the 3' terminus of said shRNA core sequence. In some embodiments, the method further comprises outputting to a user interface device, a computer readable storage medium, or a local or remote computer system; or displaying; said promoter sequence and / or said shRNA core sequence.
  • the method further comprises (a ⁇ ) selecting, prior to said selecting step (a), said first plurality of siRNA functional sequence motifs from a database of siRNA and miRNA sequences, or (a ⁇ ) selecting, prior to said selecting step (a), said first plurality of siRNA functional sequence motifs from sequences of 21U-RNAs.
  • the present invention also provides an apparatus comprising a processor, and a memory coupled to said processor and encoding one or more programs, wherein said one or more programs cause the processor to carry out a method of any embodiments in accordance with the present invention.
  • the present invention also provides a computer program product for use in conjunction with a computer having a processor and a memory connected to the processor, said computer program product comprising a computer readable storage medium having a computer program mechanism encoded thereon, wherein said computer program mechanism may be loaded into the memory of said computer and cause said computer to carry out the method of any embodiments in accordance with the present invention.
  • the invention provides a method for selecting from a plurality of different siRNAs one or more siRNAs for silencing a target gene in an organism, each of the plurality of different siRNAs targeting a different target sequence in a transcript of the target gene, the method comprising (a) ranking the plurality of different siRNAs according to positional base compositions of a corresponding targeted sequence motifs in the transcript, wherein each targeted sequence motif comprises at least a portion of the target sequence of the corresponding siRNA and/or a second sequence in a sequence region flanking the target sequence; and (b) selecting one or more siRNAs from the ranked siRNAs.
  • each sequence motif comprises the target sequence of the targeting siRNA.
  • the ranking step is carried out by (al) determining a score for each different siRNA, wherein the score is calculated using a position-specific score matrix; and (a2) ranking the plurality of different siRNAs according to the score.
  • each sequence motif is a nucleotide sequence of L nucleotides, L being an integer
  • each sequence motif is a nucleotide sequence of Z nucleotides, L being an integer
  • the score for each siRNA is calculated according to equation
  • e t and/? are respectively weights of the nucleotide at position t in the sequence motif as determined according to the position-specific score matrix and in a random sequence.
  • each sequence motif comprises the target sequence of the targeting siRNA and at least one flanking sequence.
  • each sequence motif comprises the target sequence of the targeting siRNA and a 5' flanking sequence and a 3' flanking sequence.
  • the 5' flanking sequence and the 3' flanking sequence are each a sequence of D nucleotides, D being an integer.
  • each target sequence is a sequence of 19 nucleotides, and each 5' flanking sequence and 3' flanking sequence are a sequence of 10 nucleotides.
  • each target sequence is a sequence of 19 nucleotides, and each 5' flanking sequence and 3' flanking sequence are a sequence of 50 nucleotides.
  • the one or more siRNAs consist of at least 3 siRNAs.
  • the method further comprises a step of de-overlapping, comprising selecting a plurality of siRNAs among the at least 3 siRNAs such that siRNAs in the plurality are sufficiently different in a sequence diversity measure.
  • the diversity measure is a quantifiable measure
  • the selecting in the de-overlapping step comprises selecting siRNAs having a difference in the sequence diversity measure between different selected siRNAs above a given threshold.
  • the sequence diversity measure is the overall GC content of the siRNAs.
  • the given threshold is 5%.
  • the sequence diversity measure is the distance between siRNAs along the length of the transcript sequence.
  • the threshold is 100 nucleotides.
  • the sequence diversity measure is the identity of the leading dimer of the siRNAs, wherein each of the 16 possible leading dimers is assigned a score of 1-16, respectively.
  • the threshold is 0.5.
  • the method further comprises a step of selecting one or more siRNAs based on silencing specificity, the step of selecting based on silencing specificity comprising, (i) for each of the plurality of siRNAs, predicting off-target genes of the siRNA from among a plurality of genes, wherein the off-target genes are genes other than the target gene and are directly silenced by the siRNA; (ii) ranking the plurality of siRNAs according to their respective numbers of off-target genes; and (iii) selecting one or more siRNAs for which the number of off-target genes is below a given threshold.
  • the predicting comprises (il) evaluating the sequence of each of the plurality of genes based on a predetermined siRNA sequence match pattern; and (i2) predicting the gene as an off-target gene if the gene comprise a sequence that matches the siRNA based on the sequence match pattern.
  • the step of evaluating comprises identifying an alignment of the siRNA to a sequence in a gene by a low stringency FASTA alignment.
  • each siRNA has L nucleotides in its duplex region, and the match pattern is represented by a position match position-specific score matrix
  • the step (il) comprises calculating a position match score pmScore according to equation
  • step (i2) comprises predicting the gene as an off-target gene if the position match score is greater than a given threshold.
  • the position-specific score matrix is determined by a method comprising (aa) identifying a plurality of iV siRNAs consisting of siRNAs having
  • the position-specific score matrix is obtained by a method comprising (aa) initializing the PSSM with random weights; (bb) selecting randomly a weight w ⁇ obtained in (aa); (cc) changing the value of the selected weight to generate a test psPSSM comprising the selected weight having the changed value; (dd) calculating a score for each of a plurality of siRNAs functional sequence motifs using the test PSSM according to equation
  • Score ⁇ n(w k lp k ) wherein the w* and p k are respectively weights of a nucleotide at position k in the functional sequence motif and in a random sequence; (ee) calculating correlation of the score and a metric of a characteristic of an siRNA among the plurality of siRNAs functional sequence motifs; (ff) repeating steps (cc)-(ee) for a plurality of different values of the selected weight in a given range and retain the value that corresponds to the best correlation for the selected weight; and (gg) repeating steps (bb)-(ff) for a chosen number of times; thereby determining the PSSM.
  • the method further comprises selecting the plurality of siRNA functional sequence motifs by a method comprising (i) identifying a plurality of siRNAs consisting of siRNAs having different values in the metric; (ii) identifying a plurality of siRNA functional sequence motifs each corresponding to an siRNA in the plurality of siRNAs.
  • the characteristic is silencing efficacy.
  • the plurality of N siRNAs targets a plurality of different genes having different transcript abundances in a cell.
  • step (b) is carried out by selecting one or more siRNAs having the highest scores.
  • step (b) is carried out by selecting one or more siRNAs having a score closest to a predetermined value, wherein the predetermined value is the score value corresponding to the maximum median silencing efficacy of a plurality of siRNA sequence motifs.
  • the plurality of siRNA sequence motifs are sequence motifs in transcript having abundance level of less than about 3-5 copies per cell.
  • step (b) is carried out by selecting one or more siRNAs having a score within a predetermined range, wherein the predetermined range is a score range corresponding to a plurality of siRNAs sequence motifs having a given level of silencing efficacy.
  • the silencing efficacy is above 50%, 75%, or 90% at an siRNA concentration of about 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM , preferably at 100 nM.
  • the plurality of siRNA sequence motifs are sequence motifs in transcript having abundance level of less than about 3-5 copies per cell.
  • the plurality of N siRNAs comprises at least 10,
  • siRNAs 50, 100, 200, or 500 different siRNAs.
  • the first type of siRNA consists of one or more siRNAs having silencing efficacy no less than a first threshold and the second type of siRNA consists of one or more siRNAs having silencing efficacy less than a second threshold.
  • the difference in probability is described by a sum of Gaussian curves, each of the Gaussian curves representing the difference in probability of finding a G or C at a different sequence position.
  • the first and second thresholds are both 75% at an siRNA concentration of 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, preferably at 100 nM, preferably at 100 nM.
  • the invention provides a method for selecting from a plurality of different siRNAs one or more siRNAs for silencing a target gene in an organism, each of the plurality of different siRNAs targeting a different target sequence in a transcript of the target gene, the method comprising (a) ranking the plurality of different siRNAs according to positional base composition of reverse complement sequences of sense strands of the siRNAs; and (b) selecting one or more siRNAs from the ranked siRNAs.
  • the ranking step is carried out by (al) determining a score for each different siRNA, wherein the score is calculated using a position-specific score matrix; and (a2) ranking the plurality of different siRNAs according to the score.
  • the first type of siRNA consists of one or more siRNAs having silencing efficacy no less than a first threshold and the second type of siRNA consists of one or more siRNAs having silencing efficacy less than a second threshold.
  • the difference in probability is described by a sum of Gaussian curves, each of the Gaussian curves representing the difference in probability of finding a G or C at a different sequence position.
  • the first and second thresholds are both 75% at an siRNA concentration of 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, preferably at 100 nM.
  • the invention provides a method for selecting from a plurality of different siRNAs one or more siRNAs for silencing a target gene in an organism, each of the plurality of different siRNAs targeting a different target sequence in a transcript of the target gene, the method comprising, (i) for each of the plurality of different siRNAs, predicting off-target genes of the siRNA from among a plurality of genes, wherein the off-target genes are genes other than the target gene and are directly silenced by the siRNA; (ii) ranking the plurality of different siRNAs according to the number of off-target genes; and (iii) selecting one or more siRNAs for which the number of off-target genes is below a given threshold.
  • the predicting comprises (il) evaluating the sequence of each of the plurality of genes based on a predetermined siRNA sequence match pattern; and (i2) predicting a gene as an off-target gene if the gene comprise a sequence that matches the siRNA based on the sequence match pattern.
  • pmPSSM position match position-specific score matrix
  • step (il) comprises calculating a position match score pmScore according to equation
  • step (i2) comprises predicting the gene as an off-target gene if the position match score is greater than a given threshold.
  • L is 19, and the pmPSSM is given by Table I.
  • the plurality of genes comprises all known unique genes of the organism other than the target gene.
  • the invention provides a library of siRNAs, comprising a plurality of siRNAs for each of a plurality of different genes of an organism, wherein each siRNA achieves at least 75%, at least 80%, or at least 90% silencing of its target gene.
  • the plurality of siRNAs consists of at least 3, at least 5, or at least 10 siRNAs.
  • the plurality of different genes consists of at least 10, at least 100, at least 500, at least 1,000, at least 10,000, or at least 30,000 different genes.
  • each siRNA functional sequence motif comprises the target sequence of the corresponding targeting siRNA and one or both flanking sequences of the target sequence.
  • each siRNA has M nucleotides in its duplex region, and each siRNA functional sequence motif consists of an siRNA target sequence of M nucleotides, a 5' flanking sequence of Z) / nucleotides and a 3' flanking sequence of D 2 nucleotides.
  • each siRNA has 19 nucleotides in its duplex region, and each siRNA functional sequence motif consists of an siRNA target sequence of 19 nucleotides, a 5' flanking sequence of 10 nucleotides and a 3' flanking sequence of 10 nucleotides.
  • each siRNA has 19 nucleotides in its duplex region, and each siRNA functional sequence motif consists of an siRNA target sequence of 19 nucleotides, a 5' flanking sequence of 50 nucleotides and a 3' flanking sequence of 50 nucleotides.
  • the plurality of N siRNAs each targets a gene whose transcript abundance is within a given range. In one embodiment, the range is at least about 5, 10, or 100 transcripts per cell. In another embodiment, the range is less than about 3-5 transcripts per cell.
  • the silencing threshold is 50%, 75%, or 90% at an siRNA concentration of about 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, preferably at 100 nM.
  • the plurality of N siRNAs comprises 10, 50, 100, 200, or 500 different siRNAs.
  • W k and/ty are respectively weights of a nucleotide at position k in the functional sequence motif and in a random sequence; (e) calculating correlation of the score and a metric characterizing an siRNA among the plurality of siRNAs functional sequence motifs; (f) repeating steps (c)-(e) for a plurality of different values of the selected weight in a given range and retain the value that corresponds to the best correlation for the selected weight; and (g) repeating steps (b)-(f) for a chosen number of times; thereby determining the psPSSM.
  • bsPSSM base composition position-specific score matrix
  • each siRNA functional sequence motif comprises a least a portion of the target sequence of the corresponding siRNA and/or a sequence in a sequence region flanking the siRNA target sequence
  • the method comprising (a) initializing the bsPSSM with random weights; (b) randomly selecting a weight w tJ obtained in (a); (c) changing the value of the selected weight to generate a test psPSSM comprising the selected weight having the changed value; (d) calculating a score for each of the plurality of siRNA functional sequence motifs using the test psPSSM according to equation
  • Wj 1 andp k are respectively weights of a nucleotide at position k in the functional sequence motif and in a random sequence; (e) calculating a correlation of the score and a metric of a characteristic of an siRNA among the plurality of siRNAs functional sequence motifs; (f) repeating steps (c)-(e) for a plurality of different values of the selected weight in a given range and retain the value that corresponds to the best correlation for the selected weight; and (g) repeating steps (b)-(f) for a chosen number of times; thereby determining the psPSSM.
  • each siRNA functional sequence motif comprises the target sequence of the corresponding targeting siRNA and one or both flanking sequences of the target sequence.
  • the method further comprises selecting the plurality of siRNA functional sequence motifs by a method comprising (i) identifying a plurality of siRNAs consisting of siRNAs having different values in the metric; (ii) identifying a plurality of siRNA functional sequence motifs each corresponding to an siRNA in the plurality of siRNAs.
  • each siRNA has M nucleotides in its duplex region, and each siRNA functional sequence motif consists of an siRNA target sequence of M nucleotides, a 5' flanking sequence of Dj nucleotides and a 3' flanking sequence of Z) ? nucleotides.
  • each siRNA has 19 nucleotides in its duplex region, and each siRNA functional sequence motif consists of an siRNA target sequence of 19 nucleotides, a 5' flanking sequence of 10 nucleotides and a 3' flanking sequence of 10 nucleotides.
  • each siRNA has 19 nucleotides in its duplex region, and each siRNA functional sequence motif consists of an siRNA target sequence of 19 nucleotides, a 5' flanking sequence of 50 nucleotides and a 3' flanking sequence of 50 nucleotides.
  • the metric is silencing efficacy.
  • the plurality of N siRNAs each targets a gene whose transcript abundance is within a given range.
  • the range is at least about 5, 10, or 100 transcripts per cell.
  • the range is less than about 3-5 transcripts per cell.
  • the threshold is 50%, 75%, or 90% at an siRNA concentration of about 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, preferably at 10O nM.
  • the method further comprises evaluating the psPSSM using an ROC (receiver operating characteristic) curve of the sensitivity of the psPSSM vs.
  • the non-specificity of the psPSSM curve the sensitivity of the PSSM being the proportion of true positives detected using the psPSSM as a fraction of total true positives
  • the non-specificity of the PSSM being the proportion of false positives detected using the psPSSM as a fraction of total false positives.
  • the plurality of siRNA functional sequence motifs consists of at least 50, at least 100, or at least 200 different siRNAs functional sequence motifs.
  • the method further comprises testing the psPSSM using another plurality of siRNA functional sequence motifs.
  • the position match weight matrix is given by Table I.
  • the invention also provides a method for evaluating the relative activity of the two strands of an siRNA in off-target gene silencing, comprising comparing position specific base composition of the sense strand of the siRNA and position specific base composition of the antisense strand of the siRNA or reverse complement strand of the sense strand of the siRNA, wherein the antisense strand is the guiding strand for targeting the intended target sequence.
  • the comparing is carried out by a method comprising (a) determining a score for the sense strand of the siRNA, wherein the score is calculated using a position-specific score matrix; (b) determining a score for the antisense strand of the siRNA or the reverse complement strand of the sense strand of the siRNA using the position-specific score matrix; and (c) comparing the score for the sense strand and the score for the antisense strand or the reverse complement strand of the sense strand, thereby evaluating strand preference of the siRNA.
  • the position-specific score matrix is obtained by a method comprising (a) initializing the position-specific score matrix with random weights; (b) selecting randomly a weight w y obtained in (a); (c) changing the value of the selected weight to generate a test position-specific score matrix comprising the selected weight having the changed value; (d) calculating a score for each of a plurality of siRNAs using the test position-specific score matrix according to equation
  • the w, and /7 are respectively weights of a nucleotide at position/ in the siRNA and in a random sequence; (e) calculating correlation of the score with a metric of a characteristic of an siRNA among the plurality of siRNAs; (f) repeating steps (c)-(e) for a plurality of different values of the selected weight in a given range and retain the value that corresponds to the best correlation for the selected weight; and (g) repeating steps Qo)- (f) for a chosen number of times; thereby determining the position-specific score matrix.
  • the metric is siRNA silencing efficacy.
  • the siRNA has 19 nucleotides in its duplex region.
  • the first type of siRNA consists of one or more siRNAs having silencing efficacy no less than a first threshold and the second type of siRNA consists of one or more siRNAs having silencing efficacy less than a second threshold, and the siRNA is determined as having antisense preference if the score determined in step (a) is greater than the score determined in step (b), or as having sense preference if the score determined in step (b) is greater than the score determined in step (a).
  • the difference in probability is described by a sum of Gaussian curves, each of the Gaussian curves representing the difference in probability of finding a G or C at a different sequence position.
  • the first and second thresholds are both 75% at an siRNA concentration of about 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, preferably at 100 nM.
  • the invention provides a computer system comprising a processor, and a memory coupled to the processor and encoding one or more programs, wherein the one or more programs cause the processor to carry out any one of the method of the invention.
  • the invention provides a computer program product for use in conjunction with a computer having a processor and a memory connected to the processor, the computer program product comprising a computer readable storage medium having a computer program mechanism encoded thereon, wherein the computer program mechanism may be loaded into the memory of the computer and cause the computer to carry out any one of the method of the invention.
  • FIGS. IA-C show that base composition in and around an siRNA target sequence affects the silencing efficacy of the siRNA.
  • a total of 377 siRNAs were tested by quantitative reverse transcription PCR (qRT-PCR) Taqman analysis for their ability to silence their target sequences 24hr following transfection into HeLa cells. Median target silencing was -75%.
  • This dataset was divided into two subsets, one having less than median and one having equal to or greater than median silencing ability (referred to as "bad” and "good” siRNAs, respectively). Shown here are the mean difference within a window of 5 (i.e., averaged over all 5 bases) in GC content (FIG. IA), A content (FIG. IB), and U content (FIG. 1C) between good and bad siRNAs at different relative positions on a target sequence.
  • FIGS. 2A-C show nucleotide contents of various siRNAs.
  • A GC content of good and bad siRNAs
  • B A content of good and bad siRNAs
  • C U content of good and bad siRNAs.
  • the figures show average compositions of each base. For example, 0.5 on the y- axis corresponds to an average base content of 50%.
  • FIG. 3 shows the performance of an actual siRNA base composition model used in the siRNA design method of the invention.
  • siRNA efficacy data were subdivided into two pairs of training and test sets. Different PSSMs were optimized on each of the training sets and verified on the test sets. The performance of each PSSM was evaluated by its ability to distinguish good siRNAs (true positives) and bad siRNAs (false positives) as an increasing number of siRNAs were selected from a list ranked by PSSM score. Shown are Receiver Operating Characteristics (ROC) curves demonstrating the performance of two different PSSMs on their respective training and test sets (heavy black and dotted gray lines, respectively). The expected performance of the PSSMs on randomized data is shown for comparison (i.e., no improvement in selection ability, 45 line).
  • ROC Receiver Operating Characteristics
  • FIG. 4 demonstrates the predictive ability of PSSMs on an independent experimental data set.
  • New siRNAs were designed for five genes by the standard method as described in Elbashir et al, 2001, Nature 411 :494-498, with the addition of the specificity prediction method disclosed in this application, and by the PSSM based efficacy and specificity prediction method of the invention.
  • the top three ranked siRNAs per gene were selected for each method and purchased from Dharmacon. All six siRNAs for each of the five genes were then tested for their ability to silence their target sequences. Shown is a histogram of the number of siRNAs that silence their respective target genes by a specified amount.
  • FIGS. 5A-C show mean weights of GC, A or U from the two ensembles of base composition PSSM trained and tested with siRNAs in set 1 and set 2, respectively.
  • FIG. 5A shows mean weights for GC;
  • FIG. 5B shows mean weights for A;
  • FIG. 5C shows mean weights for U.
  • siRNAs in set 1 and set 2 are shown in Table II.
  • FIG. 6 shows an example of alignments of transcripts of off-target genes to the core 19mer of an siRNA oligo sequence.
  • Off-target genes were selected from the Human 25k v2.2.1 microarray by selecting for kinetic patterns of transcript abundance consistent with direct effects of siRNA oligos.
  • the left hand column lists transcript sequence identifiers. Alignments were generated with FASTA and edited by hand. The black boxes and grey area demonstrate the higher level of sequence similarity in the 3' half of the alignment.
  • FIG. 7 shows a position match position-specific scoring matrix for predicting off- target effects.
  • the chart shows the weight associated with each position in a matrix representing the alignment between an siRNA oligo and off-target transcripts. The weight represents the probability that a match will be observed at each position / along an alignment between an siRNA oligo and an observed off-target transcript.
  • FIG. 8 shows optimization of the threshold score for predicting off-target effects of siRNAs.
  • the R values result from the correlation of number of alignments scoring above the threshold with number of observed off-target effects.
  • FIG. 9 shows a flow chart of an exemplary embodiment of the method for selecting siRNAs for use in silencing a gene.
  • FIG. 10 illustrates sequence regions that can be used for distinguishing good and bad siRNAs.
  • PSSMs were trained on chunks of sequence 10+ bases in length, from 50 bases upstream to 50 bases downstream of the siRNA 19mer, and tested on independent test sets. The performance of models trained on chunks of interest was compared with models trained on random sequences. Position 1 corresponds to the first 5' base in the duplex region of a 21 nt siRNA.
  • FIGS. 1 IA-B show curve models for PSSM.
  • 1 IA an exemplary set of curve models for PSSM.
  • 1 IB the performance of the models on training and test sets.
  • FIG. 12 illustrates an exemplary embodiment of a computer system useful for implementing the methods of the present invention.
  • FIG. 13 shows a comparison of the distribution of silencing efficacies of the siRNAs among the 30 siRNAs designed using the method of the invention (solid circles) and siRNAs designed using the standard method (open circles).
  • X-axis 1, KIF14; 2, PLK; 3, IGFlR; 4, MAPK14; 5, KIFl 1.
  • Y-axis RNA level.
  • the siRNAs designed using the standard method exhibited a broad distribution of silencing abilities, while those designed with the method of the invention show more consistent silencing within each gene, as well as across genes. A narrow distribution is very important for functional genomics with siRNAs.
  • FIGS. 14A-B show the results of a comparison of the GC content of siRNAs and their reverse complements with the GC content of bad siRNAs. The results indicate that bad siRNAs have sense strands similar to good siRNAs, while good siRNAs have sense strands similar to bad siRNAs.
  • RC reverse complement of the siRNA target sequence.
  • FIG. 15 shows that less effective siRNAs have active sense strands.
  • Strand bias of 61 siRNAs was predicted from expression profiles by the 3 '-biased method, and from comparison of the GC PSSM scores of the siRNAs and their reverse complements. Strand bias predictions were binned by siRNA silencing efficacy.
  • FIG. 16 shows that silencing efficacy relates to transcript expression level.
  • a total of 222 siRNAs (3 siRNAs per gene for 74 genes) were tested by branched DNA (bDN A) or Taqman analysis for their ability to silence their target sequences 24hr following transfection into HeLa cells.
  • Percent silencing (y-axis) was plotted as a function of transcript abundance (x-axis) measured as intensity on microarray. Shown is the median target silencing observed for 3 siRNAs per gene selected by the previous siRNA design algorithm.
  • the dependence of silencing on gene expression level, as the average of intensities from 2 array types, is shown for 74 genes.
  • TaqMan assays were used for 8 genes.
  • b-DNA data is shown for the remaining 66 genes.
  • siRNAs to poorly-expressed genes were tested by bDNA analysis for their ability to silence their target sequences. Data were divided into subsets having less than 75% silencing and equal to or greater than 75% silencing (bad and good siRNAs, respectively). Shown here is the difference in GC content between good and bad siRNAs (y-axis) at each position in the siRNA sense strand (x-axis.) The dataset includes both poorly-expressed and highly-expressed genes from 570 siRNAs selected to 33 poorly- and 41 highly-expressed genes by Tuschl rules or randomized selection. The siRNA sequences are listed in Table IV. The GC profile for good siRNAs to poorly-expressed genes (gray dotted curve) shows some similar composition preferences to good siRNAs for well-expressed genes (black curve), but also some differences.
  • FIG. 18 shows the efficacies of newly design siRNAs.
  • siRNAs were designed for 18 poorly-expressed genes by the standard method and by the new algorithm.
  • Standard pipeline selection for maximum pssm score; minimax filter for long off-target matches.
  • Improved pipeline selection for 1-3 G+C in sense 19mer bases 2-7, base 1 & 19 asymmetry, -300 ⁇ pssm score ⁇ +200, and blast matches less than 16, 200 bases on either side of the 19mer are not repeat or low-complexity sequences.
  • the top three ranked siRNAs per gene were selected for each method. All six siRNAs for each of the five genes were then tested for their ability to silence their target sequences.
  • FIG. 19 Design features of efficacious siRNAs. Studies of design criteria that correlate with siRNA silencing efficacy have revealed a number of features that predict efficacy. These include a base asymmetry at the two termini to direct the antisense (guide) strand into RISC, a U at position 10 for effective cleavage of the transcript, a low GC stretch encompassing the center and 3' end of the guide strand for enhanced cleavage, and the "seed" region at the 5' end of the antisense strand implicated in transcript binding. Gray lines above the duplex indicate sequence preferences, light gray lines below the duplex indicate functional attributes.
  • FIG. 20 shows expression vs. median silencing in 371 siRNAs. These are siRNAs from the original training set of 377 siRNAs. 6 siRNAs were not included in the analysis, as the expression level of their target gene was not available.
  • FIG. 21 shows that shRNAs and siRNAs have similar requirements for GC-content asymmetry.
  • the fractional differences in GC content preference (Y axis) of 218 shRNAs (dashed line) and 377 siRNAs (solid line) are shown for each position in the shRNA sense strand (X axis).
  • shRNAs were divided into those with greater ("good") or less (“bad") than 66% silencing, which is the median silencing for shRNAs designed by the second generation algorithm.
  • siRNAs were divided into those with greater ("good") or less (“bad”) than 75% silencing, which is the median silencing measured for siRNAs designed by a pseudorandom approach.
  • Base composition profiles are smoothed over a window of 5 bases.
  • FIG. 22 shows the preferences of siRNAs and shRNAs for individual bases at each position of the core sequence.
  • shRNAs were divided into those with greater or less than 66% silencing, the median silencing for shRNAs designed using the second generation design, and siRNAs were divided into those with greater or less than 75% silencing, the median silencing of siRNAs designed in a pseudo-random manner.
  • the fractional difference in content of each base is shown between good and bad shRNAs (dotted line) and between good and bad siRNAs (solid line) at each position in the sense strand. The strongest preference is for a G at position 1 of the sense strand.
  • FIG. 23 shows improved efficacies of shRNAs designed using the third generation algorithm.
  • shRNAs were selected by a pseudorandom design (dotted line), the second generation algorithm (dashed line), or the third generation algorithm with a requirement for G at position 1 (solid line). Shown is a histogram of the number of shRNAs that silence their target genes by a specified amount. Median silencing values were ⁇ 54% for shRNAs designed using the pseudorandom algorithm, -66% for shRNAs designed using the second generation algorithm (RSTA shRNA V2), and -75% for shRNAs designed using the third generation algorithm (RSTA shRNA V3) with the additional requirement that there is a G at position 1.
  • FIG. 1 shows improved efficacies of shRNAs designed using the third generation algorithm.
  • genes 24 shows some exemplary genes used in studies of pseudorandom shRNAs in accordance with the present invention. A dozen genes were selected based on a distribution of their overall transcript GC content (expressed as a percentage in parentheses) and of the mean log expression level as determined by microarray analysis (also in parentheses).
  • the present invention provides systems and methods for designing shRNA sequences for gene silencing in a cell through RNA interference (RNAi).
  • the cell can be a cell from an organism, a cell of an organism, a cell from a cell line, a primary cell, in vivo or ex vivo (e.g., in cell culture).
  • Each of the designed shRNA “sequences comprises an shRNA core sequence that comprises nucleotide sequence motifs joined by a loop sequence.
  • the shRNA design when transcribed and post-transcriptionally processed (e.g. , by Dicer), is predicted to yield a corresponding siRNA.
  • nucleotide sequence motifs in the shRNA designs are inverted respeats that correspond to the sense and antisense sequences of the corresponding siRNA.
  • the silencing efficacy of the corresponding siRNA is embedded in the sequence motifs of the sense or antisense strands. Accordingly, the sequence motif of one of the strands is therefore referred to as an "siRNA functional sequence motif or simply a "functional sequence motif in the present application.
  • An siRNA with a functional sequence motif causes gene silencing of a target gene by binding to an siRNA susceptible sequence motif within the target gene.
  • a sequence motif in a transcript that may be targeted by an siRNA for degradation of the transcript e.g., a sequence motif that is likely to be a highly effective siRNA targeting site is also referred to as an siRNA susceptible motif.
  • a sequence motif in a transcript of the target gene that may be less desirable for targeting by an siRNA e.g., a sequence motif that is likely to be a less effective siRNA targeting site is also referred to as an siRNA resistant motif.
  • transcription efficiency of an shRNA core sequence may be manipulated by adding flanking sequences on one or both sides of the shRNA core sequence.
  • the present invention provides systems and methods for shRNA design by separately optimizing the shRNA core sequence for silencing efficacy and the transcription sequence motif for transcription efficiency within a specific shRNA design.
  • Systems and methods for predicting silencing efficacy of a core sequence have been disclosed previously, for example, in PCT publication WO 2005/042708 A2 (hereinafter "WO 05/042708"), published on May 12, 2005.
  • an shRNA core sequence comprises the sense and antisense sequences (i.e., the functional sequence motifs) of a corresponding siRNA.
  • Sections 5.1.1 through 5.1.4 herein disclose methods and systems for designing a shRNA core sequence based on a predicted or observed silencing efficacy of a functional sequence motif that correspond to the core sequence.
  • An shRNA core sequence further comprises a loop sequence.
  • Section 5.1.5 discloses methods and system for designing sense strand and antisense strand sequences (e.g., the inverted repeats) in the shRNA core sequences using the functional sequence motifs identified in sections 5.1.1 through 5.1.4.
  • Section 5.1.5.2 further discloses exemplary methods for enhancing transcription efficiency of an shRNA sequence design and provides examples of loop sequence.
  • the invention provides a method of designing shRNA core sequences, in particular the sequences in the sense and antisense strands, by identifying the siRNA susceptible motifs for a target gene.
  • Gene silencing occurs when a strand of an siRNA binds through complementarity to an siRNA susceptible sequence motif in a target gene.
  • defining the sequence of an siRNA susceptible sequence motif sufficiently defines the functional sequence motif of an siRNA, which in turn sufficiently defines the sense strand and antisense strand sequences of an shRNA core sequence.
  • the present invention provides a method for identifying siRNA susceptible motifs in a transcript using a position-specific score matrix approach, for example, as disclosed in PCT publication No. WO 05/042708 ("hereinafter WO 05/042708").
  • WO 05/042708 provides a method for identifying off-target genes of an siRNA and for predicting specificity of an siRNA using a position-specific score matrix approach.
  • WO 05/042708 further provides a method for designing siRNAs with high silencing efficacy and specificity as well as a library of siRNAs comprising siRNAs with high silencing efficacy and specificity.
  • the present application invention further provides a method for designing shRNAs based on not only the silencing efficiencies of siRNAs produced from the shRNAs but also the transcription efficiencies of the shRNAs.
  • an shRNA core sequence is often said to target a gene through an siRNA that is produced from the shRNA core sequence.
  • the shRNA that corresponds to the shRNA is designed to target and cause degradation of a transcript of the gene.
  • a gene is also referred to as a target gene of the shRNA, and the sequence in the transcript that is acted upon by the siRNA is referred to as the target sequence.
  • a 19-nucleotide sequence in a transcript which is identical to the sequence of the 19-nucleotide sequence in the sense strand of the duplex region of the corresponding siRNA is the target sequence of the shRNA.
  • the antisense strand of the corresponding siRNA a strand that acts upon the target sequence, is also referred to as the guiding strand.
  • the antisense strand of the 19-nucleotide duplex region of the corresponding siRNA is the guiding strand.
  • features of an siRNA are often referred to with reference to its sequence, e.g., positional base composition. It will be understood that, unless specifically pointed out otherwise, such a reference is made to the sequence of the sense strand of the corresponding siRNA. It will be understood that the target sequence of an shRNA.
  • nucleotide or a sequence of nucleotides in an siRNA is often described with reference to the 5 ' or 3' end of the siRNA. It will be understood that when such a description is employed, it refers to the 5' or 3' end of the sense strand of the siRNA. It will also be understood that, when a reference to the 3' end of the siRNA is made, it refers to the 3' duplex region of the siRNA, i.e., the two nucleotides of the 3' overhang are not included in the numbering of the nucleotides.
  • shRNA design of shRNA is discussed in reference to silencing a sense strand target, i.e., transcript target sequence binding to the antisense strand of the shRNA. It will be understood by one skilled in the art that the methods of the invention are also applicable to the design of siRNA for silencing an antisense target (see, e.g., Martinez et al, 2002, Cell 1 10:563-574).
  • an siRNA functional sequence motif (or simply functional sequence motif) binds to an siRNA susceptible sequence motif in a target gene prior to RNA silencing.
  • the silencing efficacy of a functional sequence motif to an siRNA susceptible sequence motif is determined by experimental methods (e.g., methods for measuring silencing efficacies or binding coefficients) or alternatively, by methods of predicting silencing efficacies (e.g., computer algorithms based on previous experimental data, as disclosed in WO 05/042708).
  • This invention provides a method for identifying siRNA functional sequence motifs and subsequently builds a profile of siRNA functional sequence motifs using, e.g., a library of siRNAs for which silencing efficacies have been determined.
  • the sequence region of interest is scanned to identify sequences that match the profile of the functional sequence motif.
  • the profile of a functional sequence motif is represented using a position-specific score matrix (PSSM).
  • PSSM position-specific score matrix
  • a general discussion of PSSM can be found in, e.g., "Biological Sequence Analysis” by R. Durbin, S. Eddy, A. Krogh, and G. Mitchison, Cambridge Univ. Press, 1998; and Henikoff et al. , 1994, J MoI Biol. 243:574-8; each of which is hereby incorporated by reference herein in its entirety.
  • a PSSM is a sequence motif descriptor which captures the characteristics of a functional sequence motif.
  • a PSSM is used to describe sequence motifs of the invention, e.g., an siRNA susceptible or siRNA resistant motif.
  • a PSSM of an siRNA susceptible motif is also referred to as a susceptible PSSM. Accordingly, a PSSM of an siRNA resistant motif is also referred to as a resistant PSSM.
  • a position-specific score matrix is also termed a position specific scoring matrix, a position weight matrix (PWM), or a Profile, and there terms are used interchangeably herein.
  • a functional sequence motif comprises one or more sequences in an siRNA target sequence.
  • the one or more sequences in an siRNA target sequence may be a sequence at 5' end of the target sequence, a sequence at 3' end of the target sequence.
  • the one or more sequences in an siRNA target sequence may also be two stretches of sequences, one at 5' end of the target sequence and one at 3' end of the target sequence.
  • a functional sequence motif can also comprise one or more sequences in a sequence region that flanks the siRNA target sequence. Such one or more sequences can be directly adjacent to the siRNA target sequence. Such one or more sequences can also be separated from the siRNA target sequence by an intervening sequence.
  • FIG. 10 illustrates some examples of functional sequence motifs.
  • a functional sequence motif e.g., an siRNA susceptible or siRNA resistant sequence motif, comprises at least a portion of a sequence targeted by an siRNA.
  • the functional sequence motif comprises a contiguous sequence stretch of at least 7 nucleotides of the target sequence.
  • the contiguous sequence stretch is in a 3' region of the target sequence, e.g., beginning within 3 bases at the 3' end. In another embodiment, the contiguous sequence stretch is in a 5' region of the target sequence.
  • the functional sequence motif comprises a contiguous sequence stretch of at least 3, 4, 5, 6, or 7 nucleotides in a 3' region of the target sequence and comprises a contiguous sequence stretch of at least 3, 4, 5, 6, or 7 nucleotides in a 5' region of the target sequence.
  • the functional sequence motif comprises a contiguous sequence stretch of at least 11 nucleotides in a central region of the target sequence.
  • Sequence motifs comprise less than the full length of the target sequence can be used for evaluating siRNA target transcripts that exhibit only partial sequence identify to an siRNA (International Application No. PCT/US2004/015439 by Jackson et al., filed on May 17, 2004, which is incorporated herein by reference in its entirety).
  • the functional sequence motif comprises the full length siRNA target sequence.
  • the functional sequence motif may also comprise a flanking sequence. It has been discovered that the sequence of such flanking region plays a role in determining the efficacy of silencing.
  • a functional sequence motif e.g. , an siRNA susceptible or siRNA resistant sequence motif, comprises at least a portion of a sequence targeted by an siRNA and one or more sequences in one or both flanking regions.
  • a sequence motif can include an M nucleotides siRNA target sequence, a flanking sequence of Di nucleotides at one side of the siRNA target sequence and a flanking sequence of D ⁇ nucleotides at the other side of the siRNA target sequence where M, Di and D 2 are appropriate integers.
  • M 19.
  • Di, D 2 , or D is at least 5, 10, 20, 30, 50 nucleotides in length.
  • an siRNA susceptible or siRNA resistant sequence motif consists of an siRNA target sequence of 19 nucleotides and a flanking sequence of 10 nucleotides at either side of the siRNA target sequence.
  • an siRNA susceptible or siRNA resistant sequence motif consists of a 19 nucleotides siRNA target sequence and a 50 nucleotides flanking sequence at either side of the siRNA target sequence.
  • a sequence motif can include an M nucleotides siRNA target sequence, and one or more of the following: a contiguous sequence stretch of Di nucleotides flanking the 5' end of the target sequence, a contiguous sequence stretch of D 2 nucleotides flanking the 3' end of the target sequence, a contiguous sequence stretch of D 3 nucleotides which starts about 35 nucleotides upstream of the 5' end of the target sequence, a contiguous sequence stretch of D 4 nucleotides which starts about 25 nucleotides downstream of the 3' end of the target sequence, and a contiguous sequence stretch of D 5 nucleotides which starts about 60 nucleotides downstream of the 3' end of the target sequence, where Dj, D 2 , Di, D 4 , and Ds are appropriate integers.
  • each of Di, D 2, D$ t D 4, and D 5 is at least 5, 10, or 20 nucleotides in length.
  • the sequence motif include 19 nucleotides siRNA target sequence, a contiguous sequence stretch of about 10 nucleotides flanking the 5' end of the target sequence, a contiguous sequence stretch of about 10 nucleotides flanking the 3' end of the target sequence, a contiguous sequence stretch of about 10 nucleotides which starts about 35 nucleotides upstream of the 5' end of the target sequence, a contiguous sequence stretch of about 10 nucleotides which starts about 25 nucleotides downstream of the 3' end of the target sequence, and a contiguous sequence stretch of about 10 nucleotides which starts about 60 nucleotides downstream of the 3' end of the target sequence (see FIG. 10).
  • a functional sequence motif e.g., an siRNA susceptible or siRNA resistant sequence motif, comprises one or more sequences in one or both flanking regions of an siRNA target sequence but does not comprise any siRNA target sequence.
  • the functional motif comprises a contiguous sequence stretch of about 10 nucleotides flanking the 5' end of the target sequence.
  • the functional motif comprises a contiguous sequence stretch of about 10 nucleotides flanking the 3' end of the target sequence.
  • the functional motif comprises a contiguous sequence stretch of about 10 nucleotides flanking the 5' end of the target sequence and a contiguous sequence stretch of about 10 nucleotides flanking the 3' end of the target sequence.
  • the functional motif comprises a contiguous sequence stretch of about 10 nucleotides which starts about 35 nucleotides upstream of the 5' end of the target sequence. In another embodiment, the functional motif comprises a contiguous sequence stretch of about 10 nucleotides which starts about 25 nucleotides downstream of the 3' end of the target sequence. In still another embodiment, the functional motif comprises a contiguous sequence stretch of about 10 nucleotides which starts about 60 nucleotides downstream of the 3' end of the target sequence.
  • the functional motif comprises a contiguous sequence stretch of about 10 nucleotides flanking the 5' end of the target sequence, a contiguous sequence stretch of about 10 nucleotides flanking the 3' end of the target sequence, a contiguous sequence stretch of about 10 nucleotides which starts about 35 nucleotides upstream of the 5' end of the target sequence, a contiguous sequence stretch of about 10 nucleotides which starts about 25 nucleotides downstream of the 3' end of the target sequence, and a contiguous sequence stretch of about 10 nucleotides which starts about 60 nucleotides downstream of the 3' end of the target sequence.
  • a functional sequence motif can include a contiguous sequence stretch of Dy nucleotides flanking the 5' end of the target sequence, a contiguous sequence stretch of D 2 nucleotides flanking the 3' end of the target sequence, a contiguous sequence stretch of D 3 nucleotides which starts about 35 nucleotides upstream of the 5' end of the target sequence, a contiguous sequence stretch of D 4 nucleotides which starts about 25 nucleotides downstream of the 3' end of the target sequence, and a contiguous sequence stretch of D 5 nucleotides which starts about 60 nucleotides downstream of the 3' end of the target sequence, where Di, D 2 , D 3 D 4 , and D 5 are appropriate integers.
  • each of Di, D 2 , D 3 ⁇ D 4 , and Ds is at least 5, 10, or 20 nucleotides in length.
  • the characteristics of a functional sequence motif are characterized using the frequency of each of G, C, A, U(or T) observed at each position along the functional sequence motif.
  • U(or T), or sometimes simply U(T) is used to indicate nucleotide U or T.
  • the set of frequencies forms a frequency matrix, in which each element indicates the number of times that a given nucleotide has been observed at a given position.
  • a frequency matrix of a functional sequence motif can be derived or built from a set of iVsiRNA target sequences that exhibit a desired quality, e.g., a chosen level of susceptibility or resistance to siRNA silencing:
  • a functional sequence motif consists of M nucleotides siRNA target sequence, a flanking sequence of Z) / nucleotides at one side of the siRNA target sequence and a flanking sequence of D 2 nucleotides at the other side of the siRNA target sequence, L - M+ Di + D 2 .
  • the functional sequence motif consists of M nucleotides siRNA target sequence
  • a contiguous sequence stretch of D / nucleotides flanking the 5' end of the target sequence a contiguous sequence stretch of D2 nucleotides flanking the 3' end of the target sequence
  • a contiguous sequence stretch of D 3 nucleotides which starts about 35 nucleotides upstream of the 5' end of the target sequence a contiguous sequence stretch of D 4 nucleotides which starts about 25 nucleotides downstream of the 3' end of the target sequence
  • the characteristics of a functional sequence motif are characterized using a set of weights, one for each nucleotide occurring at a position in the motif.
  • the weight e l ⁇ is the probability of finding the /th nucleotide at the 7th position in the functional sequence motif.
  • the matrix is also called a probability matrix.
  • a probability matrix of a sequence motif can be derived from a frequency matrix according to equation
  • a position-specific score matrix is used to characterize a functional sequence motif.
  • the PSSM can be constructed using log likelihood values ⁇ og(e ⁇ /p, j ), where e ⁇ is the weight of finding nucleotide i at position/, and p y is the weight of finding nucleotide i at position/ in a random sequence.
  • the probability of finding the /th nucleotide at the/th position in the functional sequence motif is used as e y
  • the probability of finding nucleotide / at position/ in a random sequence is used asp, j .
  • the weight or probability p ⁇ is an "a priori" weight or probability.
  • p, ⁇ 0.25 for each possible nucleotide / e ⁇ G, C, A, U(T)) at each position /.
  • the sum of log likelihood ratios at all positions can be used as a score for evaluating if the given sequence is more or less likely to match the functional sequence motif than to match a random sequence:
  • e ⁇ and/? are respectively weights of a nucleotide at position/ in the functional sequence motif and in a random sequence. For example, if such a score is zero, the sequence has the same probability to match the functional sequence motif as to that to match a random sequence. A sequence is more likely to match the functional sequence motif if the ratio is greater than zero.
  • a PSSM with a reduced dimension when two or more different nucleotides are not to be distinguished, a PSSM with a reduced dimension can be used.
  • a PSSM has 3 sets of weights: GC-specific, A-specific and U-specific, e.g., if the base at a position is a G or a C, the natural logarithm of the ratio of the GC weight and the unbiased probability of finding a G or C at that position is used as the GC-specific weight for the position; and the natural logarithms of the position-specific A and T weights divided by the unbiased probability of respective base are used as the A- and T-specific weights for the position, respectively.
  • the log likelihood ratio score is represented by Eq. (5):
  • a PSSM has 1 set of GC-specific weights: if the base at a position is a G or a C, the natural logarithm of the ratio of the GC weight and the unbiased probability of finding a G or C at that position is used as the GC-specific weight for the position.
  • Silencing efficacies are used to validate methods for RNA interference, i.e., to assess the effect of siRNA on its target gene.
  • the silencing efficacy of a functional sequence motif in an siRNA is determined by percentage reduction of the expression level of its target gene in the presence of a given concentration of the functional sequence motif, in comparison to the expression level of the same target gene in the absence of the functional sequence motif.
  • the reduction of target gene expression level may be quantified, for example, as a reduction in transcription level, translation level or protein activity level of the target gene or a product thereof. Any suitable methods may be applied to quantify the reduction in target expression level in order to determine a silencing efficacy for the functional sequence motif.
  • Changes in transcription level e.g., endogenous mRNA level
  • qRT-PCR quantitative Reverse Transcription Polymerase Chain Reaction
  • the silencing efficacy of an siRNA functional sequence motif is determined by a branched DNA assay.
  • labeled branched DNA is mixed with a sample to be tested. If the branched DNA binds to viral RNA, then the luminescent compound will react, and the result can be measured via luminometer. See, for example, Murphy et ah, 1999, "Reproducibility and performance of the second-generation branched-DNA assay in routine quantification of human immunodeficiency virus type 1 RNA in plasma," J CHn Microbiol. 37:812-814.
  • Changes in translation level may be quantified by, for example, Western blot or other antibody-related assays, immunofluorescence (IF), enzyme-linked immunosorbent assay (ELISA), or fluorescence-activated cell sorting (FACS).
  • changes of translation level or protein expression level may be quantified by analyzing the activity of the protein product of the target gene, by using, for example, any enzymatic assay that detects changes in the activities of the protein product.
  • silencing efficacies are often expressed as a percentage in reduction of expression levels at a given siRNA concentration. For example, for a given functional sequence motif, a silencing efficacy of 90% at 100 nM siRNA means that expression of the target gene of the functional sequence motif is reduced by 90% when the siRNAs comprising the functional sequence motif are present at a concentration of 100 nM.
  • any method for determining siRNA concentration may be used to quantify the concentration of the siRNA functional sequence motif corresponding to a particular shRNA design, for example, those as disclosed in WO 05/042708 and in Section 5.5 of the instant application. It is to be understood that the concentration of the corresponding siRNA functional sequence motif of a particular shRNA design will be affected by the transcription efficiency of the shRNA design. A more detailed description of transcription efficiency is found in Section 5.1.5.3. hi some embodiments, the intracellular concentration of the siRNA functional sequence motif corresponding to a particular shRNA design may be quantified experimentally by chemical probes or tags that are associated with the transcription system (e.g., expression vector for transcribing the shRNA design).
  • EGFP Enhanced Green Fluorescence Protein
  • the concentration of an shRNA may be determined based on a predicted concentration of the corresponding siRNA functional sequence motif determined using any method known in the art, for example, those as disclosed in United States Patent Publication No. 20060166247, United States Patent Publication No. 20070149470 and in Wang et al., 2005, "Small hairpin RNAs efficiently inhibit hepatitis C IRES-mediated gene expression in human tissue culture cells and a mouse model," MoI Ther. 12:562-568.
  • siRNAs for which some quantity or quantities characterizing the siRNAs have been determined.
  • a plurality of siRNAs whose silencing efficacy has been determined can be used for determination of a PSSM of an siRNA susceptible or siRNA resistant sequence motif as disclosed WO 05/042708.
  • silencing efficacy is often used as a measure for classifying siRNAs and accordingly shRNAs with the siRNA susceptible or siRNA resistant sequence motifs embedded in the sequences of their inverted repeats.
  • the silencing efficacies of the shRNA designs directly correlate with the sequence motif of their corresponding siRNAs.
  • silencing efficacies of the corresponding siRNAs should also apply to shRNA design.
  • silencing efficacy of an siRNA is measured in the absence of other siRNAs designed to silence the target gene. It will be apparent to one skilled person in the art that the methods of the invention are equally applicable in cases where siRNAs are classified based on another measure. Such a plurality of siRNAs is also referred to as a library of siRNAs.
  • a plurality of siRNA functional motifs i.e., a sequence comprising the siRNA target sequence and the sequences in the flanking region(s) in a transcript, can be used to determine the PSSM of the functional motif.
  • the siRNA functional sequence motif consists of an siRNA target sequence of 19 nucleotides and a flanking sequence of 10 nucleotides at either side of the siRNA target sequence.
  • a library of siRNAs is often used to refer to both a library of siRNAs and a library of siRNA functional sequence motifs.
  • the efficacy of an siRNA when referred to, it refers to the efficacy of the siRNA that targets the motif.
  • the plurality of siRNAs or siRNA target motifs comprises at least 10, 50, 100, 200, 500, 1000, or 10,000 different siRNAs or siRNA target motifs.
  • each different siRNA in the plurality or library of siRNAs or siRNA functional sequence motifs can have a different level of efficacy.
  • the plurality or library of siRNAs consists of siRNAs having a chosen level of efficacy.
  • the plurality or library of siRNAs comprises siRNAs having different levels of efficacy.
  • siRNAs may be grouped into subsets, each consisting of siRNAs that have a chosen level of efficacy.
  • a PSSM of an siRNA functional motif is determining using a plurality of siRNAs having a given efficacy.
  • a plurality of N siRNAs consisting of siRNAs having a silencing efficacy above a chosen threshold is used to determine a PSSM of an siRNA susceptible motif.
  • the PSSM is determined based on the frequency of a nucleotide appeared at a position (see Section 5.1.1).
  • the chosen threshold can be 50%, 75%, 80%, or 90%.
  • a plurality of N siRNAs consisting of siRNAs having a silencing efficacy below a chosen threshold is used to determine a PSSM of an siRNA susceptible motif.
  • the chosen threshold can be 5%, 10%, 20%, 50%, 75% or 90%.
  • the PSSM has a reduced dimension with a weight for G/C.
  • a PSSM of an siRNA susceptible or siRNA resistant motif is derived or built using a classifier approach with a set of N sequences.
  • a library of siRNAs comprising siRNAs with different levels of silencing efficacies is used.
  • siRNAs in the library may be randomly grouped into subsets, each consisting of siRNAs that have different levels of silencing efficacy, one subset is used as a training set for determining a PSSM and the other is used as a testing set for validating the PSSM.
  • siRNA library Different criteria can be used to divide the existing siRNA library into training and test sets.
  • siRNA library in which a majority of siRNA oligos are designed with the standard method, which requires an AA dimer immediately before the 19mer oligo sequence, several partitions were used and more than one trained PSSMs (rather than single PSSMs) were combined to assign scores to the test oligos.
  • PSSMs training and test sets
  • the sequence motif consists of 39 bases in the transcript sequence, beginning 10 bases upstream of the 19mer siRNA target sequence and ending 10 bases downstream of the 19mer.
  • the PSSM characterizing such a sequence motif is described in Section 5.1.1.
  • the PSSM is determined by an iterative process.
  • a PSSM is initialized with random weights ⁇ e y ⁇ or ⁇ E,y ⁇ within a given search range for all bases at all positions.
  • PSSM is initialized to the smoothed mean base composition difference between good and bad siRNAs in the training set.
  • a PSSM describing a 39 nucleotide sequence motif can have 117 elements.
  • the weights are optimized by comparing the correlation of scores generated to a quantity of interest, e.g., silencing efficacy, and selecting the PSSM whose score best correspond to that quantity. Improvement in PSSM performance is scored by comparing correlation values before and after a change in weights at any one position.
  • Aggregate improvement is calculated as the difference between the final correlation and the initial correlation, hi one embodiment, for a PSSM characterizing a 39mer sequence motif, the aggregate improvement threshold after 117 cycles for termination of optimization is a difference of 0.01.
  • the weights are optimized to reflect base composition differences between good siRNAs, i.e., siRNAs having at least median silencing efficacy, and bad siRNAs, i.e., siRNAs having below median silencing efficacy, in the range of allowed values for weights.
  • the PSSM is initialized with a frequency matrix
  • the range of allowed values corresponds to the frequency matrix elements +/- 0.05.
  • the ranges of the allowed values for weights are 0.45-.55 for G/C and 0.2- 0.3 for A or U.
  • weights are allowed to vary from initial values by +/- 0.05. If an unbiased search is used, the PSSM weights can be set to random initial values within the unbiased search range described above.
  • the PSSM is determined by a random hill-climbing mutation optimization procedure.
  • one base at one position is randomly selected for optimization.
  • the 39 bases become a vector of 117 weights: 39 G/C weights, 39 A weights and 39 U weights.
  • One of these 117 weights is selected for optimization in each step, and is run through all values in the search range at that step.
  • scores for a training set of siRNAs are calculated. The correlation of these scores with the silencing efficacy of the siRNAs is then calculated. The weight for the position which generates the best correlation between the scores and silencing efficacy is retained as the new weight at that position.
  • the metric used to measure the effectiveness of the training and testing is the aggregate false detection rate (FDR) based on the ROC curve, and is computed as the average of the FDR scores of the top 33% oligos sorted by the scores given by the trained PSSM.
  • FDR aggregate false detection rate
  • those oligos with silencing levels less than the median are considered false, and those with silencing level higher than the median level are considered true.
  • the "false detection rate” is the number of false positives selected divided by the total number of true positives, measured at each ranked position in a list.
  • the false detection rate can be a function of the fraction of all siRNAs selected.
  • the area under the curve at 33% of the list selected as a single number representing performance.
  • all at-least-median siRNAs are called as "positives” and all worse-than-median siRNAs are called “negatives.” Thus, half the data are positives and the other half are “false positives.”
  • the area under the curve at 33% or even at 50% of the list selected should be 0.
  • a random ranking would cause equal numbers of true positives and false positives to be selected. This corresponds to an area under the curve of 0.17 at 33% of the list selected, or 0.25 at 50% of the list selected.
  • a plurality of PSSMs is obtained for a functional sequence motif using an siRNA training set.
  • a plurality of PSSMs is also referred to as an "ensemble" of PSSMs.
  • Each round of optimization may stop at a local optimum distinct from the global optimum.
  • the particular local optimum reached is dependent on the history of random positions selected for optimization.
  • a higher improvement threshold may not bring a PSSM optimized to a local optimum closer to the global optimum.
  • Additional runs e.g., up to 200
  • Running more than 200 optimizations was not seen to provide further enhancements in performance.
  • scoring siRNAs via the average of multiple runs is less effective than scoring candidate siRNAs on the PSSMs generated by each run and then summing the scores.
  • the PSSMs are used individually or summed to generate a composite score for each sequence match.
  • the plurality of matrices can be tested individually or as a composite on an independent set of siRNA target motifs with known silencing efficacy to evaluate the utility for identifying sequence motifs and in siRNA design.
  • the plurality of PSSM consists of at least 2, 10, 50, 100, 200, or 500 PSSMs.
  • one or more different siRNA training sets are used to obtain one or more ensemble of PSSMs.
  • These different ensembles of PSSMs may be used together in determining the predicted silencing efficacy of a sequence motif.
  • Sequence weighting methods have been used in the art to reduce redundancy and emphasize diversity in multiple sequence alignment and searching applications. Each of these methods is based on a notion of distance between a sequence and an ancestral or generalized sequence.
  • PSSMs are generated by a method which hypothesized dependency of the base composition of any one position on its neighboring positions, referred to as "curve models.”
  • curve models are generated as a sum of normal curves (i.e. , Gaussian). It will be apparent to one skilled person in the art that other suitable curve functions, e.g. , polynomials, can also be used. Each curve represents the probability of finding a particular base in a particular region. The value at each position in the summed normal curves is the weight given to that position for the base represented by the curve.
  • the weights for each base present at each position in each siRNA and its flanking sequences are then summed to generate an siRNA's score, i.e., the score is ⁇ Wj.
  • the score calculation can also be described as the dot product of the base content in the sequence with the weights in the curve model. As such, it is one way of representing the correlation of the sequence of interest with the model.
  • Curve models can be initialized to correspond to the major peaks and valleys present in the smoothed base composition difference between good and bad siRNAs, e.g., as described in FIGS. IA-C and 5A-C.
  • curve models for G/C, A and U are obtained.
  • the initial model can be set up for the 3 -peak G/C curve model as follows: Peak 1 mean: 1.5 standard deviation: 2 amplitude: 0.0455 Peak 1 mean, standard deviation and amplitude are set to correspond to the peak in the mean difference in GC content between good and bad siRNAs occurring within bases -2 - 5 of the siRNA target site in Set 1 training and test sets. Peak 2 mean: 11 standard deviation: 0.5 amplitude: 0.0337
  • Peak 2 mean standard deviation and amplitude are set to correspond to the peak in the mean difference in GC content between good and bad siRNAs occurring within bases 10- 12 of the siRNA target site in Set 1 training and test sets.
  • Peak 3 mean 18.5 standard deviation: 4 amplitude: -0.0548
  • Peak 3 mean, standard deviation and amplitude are set to correspond to the peak in the mean difference in GC content between good and bad siRNAs occurring within bases 12- 25 of the siRNA target site in Set 1 training and test sets.
  • Peak height (amplitude), center position in the sequence (mean) and width (standard deviation) of a peak in a curve model can be adjusted.
  • Curve models are optimized by adjusting the amplitude, mean and standard deviation of each peak over a preset grid of values. In one embodiment, curve models are optimized on several training sets and tested on several test sets, e.g., training sets and test sets as described in Table II. Each base - G/C, A and U (or T) - is optimized separately, and then combinations of optimized models are screened for best performance.
  • optimization criteria for curve models are: (1) the fraction of good oligos in the top 10%, 15%, 20% and 33% of the scores, (2) the false detection rate at 33% and 50% of the siRNAs selected, and (3) the correlation coefficient of siRNA silencing vs. siRNA scores used as a tiebreaker.
  • the model is trained, a grid of possible values for amplitude, mean and standard deviation of each peak is explored. The models with the top value or within the top range of values for any of the above criteria were selected and examined further.
  • G/C models are optimized with 3 or 4 peaks
  • a models are optimized with 3 peaks
  • U models are optimized with 5 peaks.
  • Exemplary ranges of parameters optimized for curve models are shown in Example 3, infra.
  • the performance of the obtained PSSM is evaluated.
  • the PSSM is evaluated using an ROC (receiver operating characteristic) curve.
  • An ROC curve is a plot of the sensitivity of a diagnostic test as a function of non-specificity.
  • An ROC curve indicates the intrinsic properties of a test's diagnostic performance and can be used to compare relative merits of competing procedures.
  • the sensitivity of a PSSM is calculated as the proportion of true positives detected as a fraction of total true positives, whereas the non-specificity of the PSSM is calculated as the proportion of false positives detected as a fraction of total false positives (see, e.g., Campbell, 1994, Statistics in Medicine 13:499-508; Metz, 1986, Investigative Radiology 21 :720-733; Gribskov et al. , 1996, Computers Chem. 20:25-33).
  • FIG. 3 shows ROC curves of the two PSSMs selected for the current best practice of the invention.
  • the performance of a PSSM is evaluated by comparing a plurality of sequence motifs identified using the PSSM with a plurality of reference sequence motifs.
  • the PSSM is used to obtain the plurality of sequence motifs by, e.g., scanning one or more transcripts and identifying sequence motifs that match the PSSM, e.g., with a score above a threshold.
  • the plurality comprises at least 3, 5, 10, 20 or 50 different sequence motifs.
  • the reference sequence motifs can be from any suitable source.
  • a plurality of reference sequence motifs is obtained using a standard method (e.g., Elbashir et al., 2001, Nature. 411 :494-8). The two pluralities are then compared using any standard method known in the art to determine if they are identical.
  • the two pluralities are compared using a Wilcoxon rank sum test.
  • a Wilcoxon rank sum test tests if two pluralities of measurements are identical (see, e.g., Snedecor and Cochran, Statistical Methods, Eighth Edition, 1989, Iowa State University Press, pp. 142-144; McClave and Sincich, 2002, Statistics, Ninth Edition, Prentice Hall, Chapter 14).
  • the Wilcoxon rank sum test can be considered a non- parametric equivalent of the unpaired t-test. It is used to test the hypothesis that two independent samples have come from the same population. Because it is non-parametric, it makes only limited assumptions about the distribution of the data. It assumes that the shape of the distribution is similar in the two groups.
  • the test ranks all the data from both groups. The smallest value is given a rank of 1 ; the second smallest is given a rank of 2, and so on. Where values are tied, they are given an average rank. The ranks for each group are added together (hence the term rank sum test). The sums of the ranks are compared with tabulated critical values to generate a p value.
  • p a function of X, Y, and ⁇ , is the probability of observing a result equal or more extreme than the one using the data (X and Y) if the null hypothesis is true.
  • the value of p indicates the significance for testing the null hypothesis that the populations generating the two independent samples, X and Y, are identical.
  • X and Y are vectors but can have different lengths, i.e., the samples can have different number of elements.
  • the alternative hypothesis is that the median of the X population is shifted from the median of the Y population by a non-zero amount, ⁇ is a given level of significance and is a scalar between zero and one. hi some embodiment, the default value of ⁇ is set to 0.05. If p is near zero, the null hypothesis may be rejected.
  • the PSSM approach of the present invention was compared to the standard method (e.g., Elbashir et al., 2001, Nature 411 :494-8) for its performance in identifying siRNAs having high efficacy.
  • the results obtained with three siRNAs selected by each method are shown in Figure 3.
  • siRNAs selected by the method using the PSSM showed better median efficacy (88% as compared to 78% for the standard method siRNA) and were more uniform in their performance.
  • the minimum efficacy was greatly improved (75% as compared to 12% for the standard method).
  • Position-specific scoring matrix approaches are the preferred method of representing siRNA functional motifs, e.g., siRNA susceptible and siRNA resistant motifs.
  • the information represented by PSSMs can also be represented by other methods which also provide weights for base-composition at particular positions. This section provides such methods for evaluating siRNA functional motifs.
  • a common method of weighting base-composition at positions in a sequence is to tally the number of a particular base or set of bases in a "window" of sequence positions. Alternatively, the tally is represented as a percentage.
  • the number of values of such a score referred to as a window score, depends on the size of the window. For example, scoring a window of size 5 for G/C content may give values of 0, 1, 2, 3, 4 or 5; or 0%, 20%, 40%, 60%, 80% or 100%.
  • thermodynamic quantities reflect the composition of all bases in the window as well as their particular order. It is readily apparent to one of skill in the art that these thermodynamic quantities directly depend on the base composition of each window, and are dominated by the G/C content of the window while showing some variation with the order of the bases.
  • the information represented by the base-composition differences e.g., in Figures IA, IB and 1C, is represented by windows of base- composition corresponding to the positions to the peaks of increased or decreased composition of a particular base(s).
  • These windows can be scored for content of the particular base(s), with increased or decreased base composition corresponding to sequences which are more or less functional or resistant for siRNA targeting. For example, a 5 -base window of increased G/C content from base -1 to base 3 relative to the siRNA 19mer duplex, and a 16-base window of decreased G/C content from base 14 to base 29 relative to the siRNA 19mer duplex, can be used to represent some of the siRNA functional motifs reflected in Figure IA.
  • the scores may be used directly as a classifier: in the example of a 5-base window, a 5-part classifier is automatically available. Scores can also be compared to a calculated or empirically derived threshold to use the window as a 2-part classifier. Windows can also be used in combination.
  • the scores of each sequence over multiple windows can be summed with or without normalization or weighting.
  • scores for each window are normalized by subtracting the mean score in a set of scores and then dividing by the standard deviation in the set of scores.
  • scores are weighted by the Pearson correlation coefficient obtained by comparing that window's score with the measured efficacy of a set of siRNAs.
  • scores are normalized, and then weighted before summation.
  • Total Percent - start position as a percentage of the length of the transcript sequence.
  • 119 bases on the transcript sequence were considered (19mer plus 50 bases downstream and 50 bases upstream).
  • Windows of sizes 3-10 were examined for each position from the beginning to the end of the 119-base chunk. The following items were counted for each window position: a. Numbers of bases: A, C, G, or U. b. Numbers of pairs of bases: M (A or C), R (A or G), W (A or U), S (C or G), Y (C or U) 3 and K (G or U). c. Numbers of various ordered dimers: AC, AT, AG, MM, RY 5 KM, SW, etc. d. The longest stretches of the above one base or two-base units.
  • the letters include the bases (A, C, G, and U) and pairs of bases (M, R, W, S, Y, and K).
  • A, C, G, and U the bases
  • M the bases
  • R, W, S, Y, and K pairs of bases
  • the structural parameters are based on the following regions.
  • the 19mer oligo proper (prefix: proper) the 20mer immediate upstream the oligo (prefix: up20) the 40mer immediate upstream the oligo the 60mer immediate upstream the oligo the 20mer immediate downstream the oligo (prefix: down20) the 40mer immediate downstream the oligo the 60mer immediate downstream the oligo
  • Parameters were normalized and weighted by the Pearson correlation coefficient of the scores with the silencing efficacy of the siRNAs examined.
  • Various methods were used to select the parameters with the greatest predictive power for siRNA efficacy; the various methods agreed on the selection 1750 parameters. 1190 of these are window- based base composition parameters, 559 are motif-based base composition parameters, and only 1 structural parameter was selected. No other parameters were selected.
  • siRNA functional motifs e.g., siRNA susceptible or siRNA resistant motifs
  • an siRNA functional motif e.g. , siRNA susceptible or siRNA resistant motif
  • siRNA susceptible or siRNA resistant motif can be understood as a loose consensus sequence for a family of distantly related sequences - e.g. the family of functional siRNA target sites. Scoring sequences for similarity to a family consensus is well known in the art. More detailed discussion may be found in Gribskov, et al, 1987, "Profile analysis: detection of distantly related proteins," Proc. Natl. Acad. Sci.
  • a difference between consensus profiles and PSSMs as the term is used in this text is that spacing can be flexible in consensus profiles: discontinuous portions of an siRNA functional motif, e.g., siRNA susceptible or siRNA resistant motif can be found at varying distances to each other, with insertions or deletions permitted and scored as bases are.
  • siRNA functional motif e.g., siRNA susceptible or siRNA resistant motif
  • Profile hidden Markov models are statistical models which also represent the consensus of a family of sequences. Krogh and colleagues (Krogh et al, 1994, Hidden Markov models in computational biology: Applications to protein modeling. J. MoI Biol. 235: 1501-1531) applied HMM techniques to modeling sequence profiles, adopting techniques from speech recognition studies (Rabiner, 1989, A tutorial on hidden Markov models and selected applications to speech recognition. Proc. IEEE 77:257-286).
  • the use of hidden Markov models for analysis of biological sequences is now well known in the art and applications for hidden Markov model calculation are readily available; for example, the program HMMER from the HHMI Janelia Farm Research Campus in Ashburn, Virginia.
  • Profile hidden Markov models differ from consensus profiles as described above in that profile hidden Markov models have a formal probabilistic basis for setting the weights for each base, insertion or deletion at each position.
  • Hidden Markov models can also perform the alignment of unknown sequences for discovery of motifs as well as determining position-specific weights for said motifs, while consensus profiles are generally derived from previously aligned sequences. Consensus profiles and profile hidden Markov models can assume that the base composition at a particular position is independent of the base composition of all other positions. This is similar to the random-hill-climbing PSSMs of this invention but distinct from the windows and curve model PSSMs.
  • Markov models can be used as fixed-order Markov chains and interpolated Markov models.
  • Salzberg and colleagues applied interpolated Markov models to finding genes in microbial genomes as an improvement over fixed- order Markov chains (Salzberg et al., 1998, Nucl. Acids Res. 26: 544-548).
  • a fixed-order Markov chain predicts each base of a sequence as a function of a fixed number of bases preceding that position. The number of preceding bases used to predict the next is known as the order of the Markov chain.
  • Interpolated Markov models use a flexible number of preceding bases to predict the base composition at a particular position. This permits training on smaller sequence sets. Sufficient predictive data may be available for n-mers of various lengths in a training set such that some predictions of succeeding bases can be made, while insufficient data may be available for all oligomers at any fixed length.
  • Interpolated Markov models thus have more freedom to use preferable longer oligomers for prediction than fixed-order Markov chains, when said long oligomers are sufficiently frequent in the training set.
  • Interpolated Markov models employ a weighted combination of probabilities from a plurality of oligomer lengths for classification of each base.
  • Fixed-order Markov chains and interpolated Markov models can represent siRNA functional motifs, e.g., siRNA susceptible or siRNA resistant motifs in terms of the dependency of the base-composition at a particular position on the composition of the preceding positions. An interpolated Markov model building process will discover the oligomers most predictive of siRNA functional or nonfunctional motifs.
  • Neural networks are also employed to score sequences for similarity to a family of sequences.
  • a neural network is a statistical analysis tool used to build a model through an iterative learning process. The trained network will then perform a classification task, dependent upon the desired output and the training input initially associated with that output.
  • a neural network program or computational device is supplied with a training set of sequences and sets up a state representing those sequences. The neural network is then tested for performance on a test set of sequences.
  • Neural networks can be used to predict and model siRNA functional motifs, e.g., siRNA susceptible and siRNA resistant motifs.
  • a disadvantage of neural networks is that the actual sequence features of a motif can be difficult or impossible to determine from examination of the state of the trained network.
  • the invention provides a method for identifying one or more sequence motifs in a transcript which are siRNA-susceptible or siRNA-resistant motifs.
  • the corresponding functional or unfunctional siRNAs are thereby also provided by the method.
  • the sequence region of interest is scanned to identify sequences that match the profile of a functional motif.
  • a plurality of possible siRNA sequence motifs comprises siRNA sequence motifs tiled across the region at steps of a predetermined base intervals are evaluated to identify sequences that matched the profile. In a preferred embodiment, steps of 1, 5, 10, 15, or 19 base intervals are used.
  • the entire transcript sequence is scanned.
  • a score is calculated for each different sequence motif using a PSSM as described in Sections 5.1.1 through 5.1.3.
  • the sequences are then ranked according to the score.
  • One or more sequences are then selected from the rank list.
  • siRNA sequence motifs having the highest scores are selected as siRNA-susceptible motifs.
  • siRNA sequence motifs having the lowest scores are selected as siRNA resistant motifs. It has been discovered that the correlation between silencing efficacy and the base composition profiles of siRNA functional motifs may depend on one or more factors, e.g., the abundance of the target transcript.
  • siRNA functional motifs having high GC content asymmetry at the two ends of the target sequence and having high GC content in the sequence regions flanking the target sequence have lower silencing efficacy than siRNA functional motifs having moderate GC content asymmetry at the two ends of the target sequence and low GC content in the flanking regions.
  • siRNA functional motifs having moderate GC content asymmetry at the two ends of the target sequence and low GC content in the flanking regions have lower silencing efficacy than siRNA functional motifs having moderate GC content asymmetry at the two ends of the target sequence and low GC content in the flanking regions.
  • silencing efficacy of a particular siRNA functional motif is a result of the interplay of a number of processes, including RISC formation and siRNA duplex unwinding, diffusion of the RISC and target mRNA, reaction of the RISC/target complex, which may include diffusion of the RISC along the target mRNA, cleavage reaction, and products dissociation, etc.
  • the abundance of the transcript, the base composition profile of the siRNA, the base composition profile of the target sequence and flanking sequences, and the concentration of the siRNA and RISC in a cell may all affect silencing efficacy.
  • siRNAs may be designed based on criteria that take one or more of such features into account. For example, bases near the 5' end of the guide strand are implicated in transcript binding (both on- and off-target transcripts), and have been shown to be sufficient for target RNA- binding energy.
  • Weaker base pairing at the 5' end of the antisense strand (3' end of the duplex) encourages preferential interaction of the antisense strand with RISC, e.g., by facilitating unwinding of the siRNA duplex by a 5 '-3' helicase component of RISC.
  • a preference for U at position 10 of the sense strand of an siRNA has been associated with improved cleavage efficiency by RISC as it is in most endonucleases.
  • the invention provides a method of identifying siRNA sequence motifs (and thus siRNAs) by obtaining siRNAs that have an optimal sequence composition in one or more sequence regions such that these siRNAs are optimal in one or more the siRNA functional processes.
  • the method comprises identifying siRNA sequence motifs whose overall sequence and/or different sequence regions have desired composition profiles.
  • the method can be used to identify siRNAs motifs that have desired sequence composition in a particular region, thus are optimized for one functional process.
  • the method can also be used to identify siRNAs that have desired sequence composition in a number of regions, thus are optimized for a number of functional processes.
  • a single siRNA functional profile e.g., a profile as represented by a set of PSSMs
  • a single siRNA functional profile is obtained, e.g. , by training with silencing efficacy data of a plurality of siRNAs that target genes having different transcript abundances using a method described in Section 5.1.2 or Section 5.1.3., and is used to evaluate siRNA sequence motifs in gene transcripts having abundances in all ranges.
  • the siRNA sequence motifs in gene transcripts having abundances in any range are evaluated based on the degree of similarity of their sequence base composition profiles to the profile or profiles represented by the set of PSSMs.
  • the PSSM scores of siRNA functional motifs for a gene of interest are obtained by a method described in Section 5.1.1.
  • a predetermined reference value or reference range of values of the PSSM score is determined based on siRNAs that target genes having expression levels in different ranges. Methods for determining the reference value or range of reference value is described below. siRNA functional motifs in a particular gene are then ranked based on the closeness of their scores to the predetermined reference value or within the reference range. One or more siRNAs having scores closest to the predetermined value or within the reference range are then selected.
  • a predetermined reference value of the PSSM score or a reference range of the PSSM scores is used for genes having expression levels in a given range. The reference value or the reference range is determined based on siRNAs that target genes having expression levels in the range. siRNA functional motifs in a particular gene are then ranked based on the closeness of their scores to the predetermined reference value or within the reference range. One or more siRNAs having scores closest to the predetermined value or within the reference range are then selected.
  • the reference value or the reference range can be determined in various ways.
  • correlation of PSSM scores of a plurality of siRNAs having one or more features, e.g., having particular efficiency in one or more siRNA functional processes, with silencing efficacy is evaluated.
  • the feature is that the plurality of siRNAs targets poorly-expressed genes.
  • the value of the score corresponding to maximum median silencing is used as the reference value.
  • the reference value is 0.
  • One or more siRNAs having PSSM scores the closest to the reference score are selected.
  • the range of scores corresponding to siRNAs having a given level of silencing efficacy is used as the range for the reference values.
  • effective siRNAs are found to have scores between -300 and +200 as long as the GC content in bases 2-7 is controlled.
  • a reference value of between -300 and +200 is used.
  • One or more siRNAs having PSSM scores within the range are selected.
  • a particular score range within the range of PSSM scores of the plurality of siRNAs having one or more features, e.g., having particular efficiency in one or more siRNA functional processes, is used as the range of the reference value.
  • the feature is that the plurality of siRNAs targets poorly-expressed genes.
  • a certain percentile in the range of PSSM scores is used as the range of the reference value, e.g., 90%, 80%, 70%, or 60%.
  • the combined PSSM score range in the training set has a maximum of 200, with 97% of the scores being 0 or below and 60% of the scores are below -300.
  • the plurality of sets consists of the two sets of PSSMs described previously.
  • the two sets of PSSMs differ in the base composition preferred for siRNAs, in particular with respect to the GC content of the 19mer and flanking sequences. With a combined score of 0, the PSSM sets are in balance in their preference for the siRNA.
  • the siRNA sequence motifs are also ranked according to GC content at positions corresponding to positions 2-7 of the corresponding siRNAs, and one or more siRNA sequence motifs that have a GC content approximately 0.15 to 0.5 (corresponding to 1-3 G or C) in the region are selected.
  • siRNA sequence motifs having a G or C at the position corresponding to position 1 of the corresponding 19mer siRNA and a A or T at the position corresponding to position 19 of the corresponding 19mer siRNA are selected.
  • siRNAs motifs in which 200 bases on either side of the 19mer target region are not repeat or low-complexity sequences are selected.
  • the siRNA sequence motifs selected in the following manner (1) they are first ranked according to GC content at positions corresponding to positions 2-7 of the corresponding siRNAs, and one or more siRNA sequence motifs that have a GC content approximately 0.15 to 0.5 (corresponding to 1-3 G or C) in the region are selected; (2) next, siRNA sequence motifs having a G or C at the position corresponding to position 1 of the corresponding 19mer siRNA and a A or T at the position corresponding to position 19 of the corresponding 19mer siRNA are selected; (3) siRNAs having PSSM scores in the range of -300 to 200 or most close to 0 are then selected; (4) number of off-target BLAST match less than 16 are then selected; and (5) siRNAs motifs in which 200 bases on either side of the 19mer target region are not repeat or low-complexity sequences are selected.
  • a reference value or reference range for each of a plurality of different abundance ranges is determined. Selection of siRNA functional motifs in a gene of interest is achieved by using the appropriate reference value or reference range for the abundance range in which the gene of interest falls.
  • the plurality of different abundance ranges consists of two ranges: below about 3-5 copies per cell, corresponding to poorly-expressed genes, and above 5 copies per cell, corresponding to highly-expressed genes.
  • the reference value or reference range can be determined for each abundance range using any one of the methods described above.
  • a plurality of siRNA functional motif profiles are determined for a plurality of different transcript abundance ranges. Each such profile is determined based on silencing efficacy data of siRNAs that target genes having expression levels in a given range, /. e. , genes whose transcript abundances fall within a given range, using a method described in Sections 5.1.2 and 5.1.3., supra.
  • a set of one or more PSSMs for genes having expression levels in a given range are trained as described in Section 5.1.2., using siRNAs that target genes having expression levels in the range.
  • the PSSMs are then used for identifying siRNA functional motifs in a target gene whose expression level falls in the range, e.g., by ranking according to the PSSM scores obtained using a method described in Section 5.1.1.
  • the transcript abundance ranges are divided into two ranges: below about 3-5 copies per cell, corresponding to poorly-expressed genes, and above 5 copies per cell, corresponding to highly-expressed genes. Two sets of PSSMs are obtained, one for each abundance range.
  • siRNA functional motifs in a gene of interest can be identified using the set of PSSMs that is appropriate for the abundance of the gene of interest.
  • a plurality of siRNA functional motif profiles are determined for a plurality of different siRNA concentration ranges. Each such profile can be determined based on silencing efficacy data of different concentration of siRNAs targeting genes having a different expression level or having an expression level in a different range. In one embodiment, such profiles are determined for transcripts having a given abundance or having a abundance within a range of abundances.
  • Each such profile can be determined based on silencing efficacy data of different concentration of siRNAs targeting genes having the expression level or having an expression level in the range.
  • one or more PSSMs for a given siRNA concentration range are trained based on silencing efficacy data of siRNAs having a concentration in the range. The PSSMs can then be used for selecting siRNAs that have high efficiency at a concentration that falls in the concentration range.
  • the transcript abundance ranges is selected to be below 5 copies per cell. In another embodiment, the transcript abundance ranges is selected to be above 5 copies per cell. The invention thus provides a method for selecting one or more siRNA functional motifs for targeting by siRNAs of a given concentration.
  • the methods can be used for identifying one or more siRNA functional motifs that can be targeted by siRNAs of a given concentration with desired silencing efficacy.
  • the given concentration is preferably in the low nanomolar to sub-nanomolar range, more preferably in the picomolar range.
  • the given concentration is 50 nmol, 20 nmol, 10 nmol, 5 nmol, 1 nmol, 0.5 nmol, 0.1 nmol, 0.05 nmol, or 0.01 nmol.
  • the desired silencing efficacy is at least 50%, 75%, 90%, or 99% under a given concentration.
  • Such methods are particularly useful for designing therapeutic siRNAs. For therapeutic uses, it is often desirable to identify siRNAs that can silence a target gene with high efficacy at sub-nanomolar to picomolar concentrations. The invention thus also provides a method for design of therapeutic siRNAs.
  • a method for determining if a gene is suitable for targeting by a therapeutic siRNA is known; see for example, WO 05/042708.
  • the desired siRNA concentration and the desired silencing efficacy are first determined.
  • a plurality of possible siRNA sequence motifs in the transcript of the gene is evaluated using a method of this invention.
  • One or more siRNA sequence motifs that exhibit the highest efficacy, e.g., having PSSM scores satisfying the above described criterion or criteria, are identified.
  • the gene is determined as suitable for targeting by a therapeutic siRNA if the one or more siRNA sequence motifs can be targeted by the corresponding siRNAs with silencing efficacy above or equal to the desired efficacy.
  • the plurality of possible siRNA sequence motifs comprises siRNA sequence motifs that span or are tiled across a part of or the entire transcript at steps of a predetermined base intervals, e.g. at steps of 1, 5, 10, 15, or 19 base intervals.
  • successive overlapping siRNA sequence motifs are tiled across the entire transcript sequence.
  • successive overlapping siRNA sequence motifs tiled across a region of or the entire transcript sequence at steps of 1 base intervals.
  • Gene silencing can also be achieved using a cell-expressed short hairpin RNA (shRNA), e.g., by introducing into a cell a DNA construct which can be transcribed to express an shRNA (see, e.g., Paddison et al, 2002, Genes Dev. 16:948-958; Brummelkamp et al, 2002, Science 296:550-553; Sui et al, 2002, Proc. Natl. Acad. ScL USA 99:5515-5520; Ventura et al, 2004, Proc. Natl. Acad.
  • shRNA short hairpin RNA
  • siRNA when an siRNA is to be produced intracellularly from an shRNA, by way of example, a DNA fragment encoding an shRNA core sequence can be inserted into an expression vector, delivered into a cell where it is subsequently transported into the nucleus of the cell, for example, a human liver cell. Once inside the nucleus, a promoter sequence embedded in the expression vector is recognized by an RNA polymerase, for example, human RNA polymerase II (Pol II) or human RNA polymerase III.
  • RNA polymerase for example, human RNA polymerase II (Pol II) or human RNA polymerase III.
  • an shRNA core sequence comprises two inverted repeats that are separated by an intervening loop region.
  • the shRNA core sequence forms a hairpin like structure, hence the name short hairpin RNA, with the inverted repeats as the duplex stem connected at one end by the loop sequence.
  • processing enzymes for example Dicer, cleave the loop ends to convert shRNA into functional siRNA with gene silencing capacity.
  • the inverted repeats of the shRNA core sequence form the sense and antisense strands of the functional siRNA.
  • a functional siRNA may be produced.
  • the functional siRNA is accordingly termed a corresponding siRNA sequence to the original shRNA core sequence.
  • the overall silencing efficacy of an shRNA core sequence i.e., the silencing efficacy of a cell-expressed shRNA
  • the silencing efficacy of a cell-expressed shRNA is determined by an interplay between the silencing efficacy of the corresponding siRNA (i.e., the siRNA produced from the shRNA core sequence) and the transcription efficiency of the shRNA.
  • a G at the first or second 5' nucleotide of the sense strand provides the best transcription level under an RNA polymerase III promoter.
  • the methods, computer systems, and computer program products of the present invention take advantage of the discovery by selecting and using siRNA sequence motifs with high silencing efficacy as the antisense strand sequences of shRNA core sequences.
  • the transcription efficiency of the shRNA core sequences may be optimized to further enhancing the silencing efficacy of the shRNA core sequences.
  • the invention provides that the inverted repeat sequences in an shRNA core sequence may be designed based on a predicted silencing efficacy for the corresponding siRNA of the original shRNA core sequence.
  • sequences for the inverted repeats are designed in accordance with the methods described in sections 5.1.1 through 5.1.4. .
  • the present invention provides a method of identifying a sequence motif in a target transcript which may be targeted by a cell expressed shRNA, e.g., a sequence motif that is likely to be an effective shRNA targeting site.
  • a sequence motif is also referred to as an shRNA susceptible motif.
  • the method can also be used for identifying a sequence motif in a transcript which may be less desirable for targeting by an shRNA, e.g., a sequence motif that is likely to be a less effective shRNA targeting site.
  • Such a sequence motif is also referred to as an shRNA resistant motif.
  • an shRNA core sequence which produces a corresponding siRNA that binds to an shRNA susceptible motif likely exhibits high silencing efficacy against the target transcript.
  • the silencing efficacy of an shRNA is embedded in the inverted repeats, which are designed in accordance with the methods for identifying an siRNA susceptible motif, as disclosed in sections 5.1.1 through 5.1.4.
  • the silencing efficacy of an shRNA core sequence may be predicted based on a predicted or observed silencing efficacy of its corresponding siRNA, for example, using methods as disclosed in WO 05/042708.
  • the silencing efficacies of the corresponding siRNAs may be based on known experimental data, for example, by selecting the siRNA sequences with previously determined silencing efficacies from an siRNA database.
  • siRNA or miRNA databases or libraries include siRNAdb hosted by Center for Genomics and Bioinformatics, Karolinska Institutet in Sweden (see Chalk et al, 2005, Nucleic Acids Research 33 (Database Issue):D131-D134); HuSiDa (Human siRNA Database) hosted by Department of Pediatrics, Laboratory of Molecular Biology at Humboldt-University in Berlin, Germany; and the miRBase hosted by the Sanger Institute in Cambridge United Kingdom (see, for example, Griffiths- Jones et al., 2006, "miRBase: microRNA sequences, targets and gene nomenclature," Nucleic Acids Research, 34 (Database Issue):D140-D144; Griffiths- Jones, 2004, "The microRNA Registry,” Nucleic Acids Research, 32
  • siRNA functional sequence motifs are selected based on a predicted silencing efficacy value that is greater than a pre-determined threshold value.
  • all siRNA functional sequence motifs with predicted silencing efficacy values above the pre-determined value will be separated into two sub-groups based on the composition of the last nucleotide residue in the sequence motif.
  • the first sub-group comprises all siRNA functional sequence motifs that end with a Cytosine (C) at the 3 '-terminus.
  • an siRNA functional sequence motif will be selected from the first sub-group as the antisense strand sequence of the shRNA core sequence.
  • the sense sequence will be created as the inverted repeat to the antisense strand, thereby ensuring that the sense strand will start with a Guanine (G) at the 5 '-terminus.
  • the second sub-group comprises those siRNA functional sequence motifs that do not end with a Cytosine (C) at the 3 '-terminus.
  • an siRNA functional sequence motif will be selected from the second sub- group. The last nucleotide at the 3 '-terminus of the selected siRNA functional sequence motif will be replaced with a Cytosine (C) before it is used as the antisense strand of an shRNA core sequence.
  • the sense sequence will be created as the inverted repeat to the antisense strand, thereby ensuring that the sense strand will start with a Guanine (G) at the 5 '-terminus.
  • siRNA functional sequence motif will be selected from any of the siRNA functional sequence motifs that have silencing efficacies above a predetermined value.
  • a Cytosine (C) will be added to the 3' terminus of the selected siRNA functional sequence motif before it is used as the antisense strand of an shRNA core sequence.
  • the sense sequence will be created as the inverted repeat to the antisense strand, thereby ensuring that the sense strand will start with a Guanine (G) at the 5'- terminus.
  • shRNA core sequences that are based on siRNA functional sequence motifs or miRNA sequence motifs. Additional small RNA sequence motifs which exhibit RNA silencing may also be used in constructing the shRNA core sequences.
  • the new classes of small RNAs other than siRNAs and miRNAs e.g., 21U-RNAs
  • 21U-RNAs examples of 21U-RNAs in C. elegans can be found in Ruby et al, 2006, "Large-scale Sequencing Reveals 21U-RNAs and Additional microRNAs and Endogenous siRNAs in C. elegans," Cell 127: 1193-1207.
  • the silencing efficacy of an shRNA core sequence can be actually measured or predicted.
  • the silencing efficacy of an siRNA produced from an shRNA core sequence can be predicted using any of the methods described in Sections 5.1.1 to 5.1.4.
  • the silencing efficacy of a corresponding siRNA can be evaluated based on positional base composition of a targeted sequence motif in the transcript, e.g., a targeted sequence motif that comprises at least a portion of the target sequence of the siRNA and/or a second sequence in a sequence region within about 200 nucleotides from said target sequence in the RNA transcript.
  • the targeted sequence motif comprises at least a portion of the target sequence of the siRNA and/or a second sequence in a sequence region within about 200 nucleotides from said target sequence in the RNA transcript.
  • the targeted sequence motif comprises the target sequence of the siRNA in the transcript.
  • the length of the duplex stems of the corresponding siRNAs varies from 19 nt to 21 nt.
  • each of the inverted repeats form the corresponding siRNAs accordingly contain from 21 nt to 23 nt.
  • longer inverted repeats may be use to construct the shRNAs core sequences, for example, Kim et ⁇ l. , have used as long as 27 nt to construct double stranded RNAs (dsRNAs) for gene silencing purpose. See, Kim, et al, 2005, “Synthetic dsRNA Dicer substrates enhance RNAi potency and efficacy," Nature Biotechnology 23:222-226, which is hereby incorporated by reference herein by reference in its entirety.
  • the loop region that connects the two inverted repeats in an shRNA core sequence varies from three nucleotides to over 20 nucleotides, for example over 25 nucleotides or even longer.
  • a loop consists of nine nucleotides. Loop sequences that facilitate hairpin formation have been reported in the literatures. Examples of these loop sequences include but are not limited ATG, CCC, TTCG, CCACC, CTCGAG, AAGCTT, CCACACC, TTCAAGAGA. More details on loop sequence selection and design can be found in Sui, et al , 2002, "A DNA vector- based RNAi technology to suppress gene expression in mammalian cells," Proc. Natl. Acad.
  • the transcription efficiency of the shRNA can be predicted based on the sequence characteristics preferred by the cellular mechanism employed for the transcription of the shRNA from its encoding DNA construct (e.g., an shRNA design).
  • the transcription efficiency of an shRNA is affected by the identity of one or more nucleotides at the 5' end of said shRNA or by the context into which the shRNA is embedded.
  • an shRNA can be expressed as a simple hairpin stem loop structure or an shRNA can be inserted into a microRNA-like context such that flanking sequence derived from natural microRNAs is expressed upstream and downstream of the shRNA itself.
  • the flanking sequence can be derived from human microRNA 30 (mir30).
  • mir30 may be found in Stegmeier et al., 2005, Proc. Natl. Acad. Sci. USA 102:13212-13217; Miyagishi et al, 2004, The Journal of Gene Medicine 6:715-723; and Shin et al, 2006, Proc. Natl. Acad. Sci. USA 103: 13759-13764; each of which is hereby incorporated by reference herein in its entirety.
  • the present invention also provides that a G at the first or second 5' nucleotide of the sense strand enhances silencing efficacy of a corresponding siRNA, possibly through enhancing the transcriptional efficiency of the shRNA core sequence from which the corresponding siRNA is produced.
  • the invention provides a method for identifying an shRNA susceptible sequence motif in a transcript of a target gene.
  • the method involves identifying one or more sequence motifs that have a high silencing efficacy, e.g., using any of the methods described in Sections 5.1.1 to 5.1.4.
  • the method can involves identifying one or more sequence motifs for the inverted repeats that have a silencing efficacy above a predetermined threshold.
  • the threshold can be a rank based threshold.
  • a plurality of candidate sequence motifs shRNA inverted repeats is ranked according to their predicted silencing efficacies.
  • the candidate target sequence motifs having a silencing efficacy above a threshold percentile, e.g., 5%, 10% or 20%, in the rank list are selected.
  • a sequence motif corresponding to an shRNA having a GY (Y is any nucleotide) or XG (X is A or T) at the first and second nucleotide positions is identified as an shRNA susceptible sequence motif.
  • the overall silencing efficacy of such shRNAs from highest to lowest is as follows: for Y, G>T>A>C; for X, T>A.
  • shRNA susceptible sequence motifs and the corresponding shRNAs can be selected according to these orders for desired overall silencing efficacy.
  • a sequence motif corresponding to an shRNA having G at the first nucleotide position at the 5' end is identified as an shRNA susceptible sequence motif.
  • a sequence motif corresponding to an shRNA having a G at the second nucleotide position at the 5' end and not having a C at the first nucleotide position is identified as an shRNA susceptible sequence motif.
  • a sequence motif corresponding to an shRNA having a G at the first nucleotide position at the 5' end of said shRNA, and a G at the second nucleotide position at the 5' end is identified as an shRNA susceptible sequence motif.
  • a sequence motif corresponding to an shRNA having a G at the first nucleotide position at the 5' end, and a T at the second nucleotide position at the 5' end is identified as an shRNA susceptible sequence motif.
  • the method can also be used for identifying shRNAs for silencing a gene.
  • the shRNAs can be selected according to the overall silencing efficacies of their target shRNA functional sequence motifs by the above described methods.
  • the invention provides a method for designing an shRNA for silencing a target gene.
  • the method involves identifying one or more shRNAs whose corresponding siRNAs have silencing efficiencies above a selected threshold value, e.g., using any of the methods described in Sections 5.1.1 to 5.1.4.
  • the threshold can be a rank based threshold.
  • a plurality of candidate shRNA target sequence motifs are ranked according to their silencing efficacy.
  • the shRNA target sequence motifs having a silencing efficacy above a threshold percentile, e.g., 5%, 10% or 20%, in the rank list are selected.
  • One or more nucleotides at the 5' end of these shRNAs are examined and replaced with different nucleotides such that the modified shRNAs have enhanced overall silencing efficacy.
  • the first and second nucleotide positions at the 5' end are modified to be GY (Y is any nucleotide) or XG (X is A or T).
  • the overall silencing efficacy of such shRNAs from highest to lowest is as follows: for Y, G>T>A>C; for X, T>A.
  • shRNAs having desired silencing efficiencies can be designed according to these orders.
  • the nucleotide at the first nucleotide position at the 5' end in the identified one or more shRNAs is replaced with a G, if it is not a G.
  • the nucleotide at the second position of the 5' end in the identified one or more shRNAs is replaced with a G, if it is not a G, and the nucleotide at the first position of the 5' end is replaced such that it is not a C.
  • the replacement of the one or more nucleotides creates mismatches between the produced siRNA and its target sequence in the transcript. However, it has been found that such mismatches do not reduce the silencing efficacy of the siRNA.
  • the shRNA design containing the nucleotide sequence encoding the shRNA core sequence is under the control of a mammalian promoter (e.g., a human or mouse Hl promoter, a human or mouse U6 promoter, or a human U2 promoter) for transcription by a RNA polymerase III.
  • a mammalian promoter e.g., a human or mouse Hl promoter, a human or mouse U6 promoter, or a human U2 promoter
  • the DNA construct containing the nucleotide sequence encoding the shRNA is under the control of a RNA polymerase II (Pol II) promoter, e.g., a promoter containing a TATA upstream of the shRNA encoding sequence, for transcription by a RNA polymerase II.
  • Poly II RNA polymerase II
  • the invention provides a method for designing a DNA construct that can be introduced into a cell of an organism to produce a short hairpin RNA (shRNA) in the cell.
  • shRNA short hairpin RNA
  • the method involves evaluating the silencing efficacies of a plurality of different siRNAs, each of which targets a different target sequence in a transcript of the target gene, and selecting an siRNA that has a desired silencing efficacy, e.g., an siRNA that has a predicted silencing efficacy above a predetermined threshold.
  • the selected siRNA sequence is then filtered to eliminate the siRNA sequence motifs that do not contain a Guanine (G) at the 5' terminal position.
  • G Guanine
  • Any of the methods described in this Section and Sections 5.1.1 to 5.1.4 can be used for this purpose.
  • siRNA sequence motifs selected after the filtering step can be used as construct a shRNA core sequence, for example, as the antisense strand.
  • the shRNA core sequence can be further designed and constructed by creating a sense strand based on the antisense strand and linking the two strands of the siRNA with a loop sequence. Upon transcription, the shRNA core sequence can produce an siRNA with a functional sequence motif corresponding to the selected siRNA.
  • an shRNA design can be designed and constructed by fusing the shRNA core sequence with a promoter and at least one flanking nucleotide sequences.
  • the shRNA design comprises in 5' to 3' order (i) a promoter, (ii) the optional first flanking nucleotide sequence, (iii) the nucleotide sequence encoding the shRNA, and (iv) the optional second flanking nucleotide sequence.
  • the optional first flanking nucleotide sequence in the shRNA design encodes a 5' single-stranded flanking RNA sequence
  • the optional second flanking nucleotide sequence in the shRNA design encodes a 3' single-stranded flanking RNA sequence.
  • the shRNA design when transcribed under the control of the promoter, produces an RNA hairpin (hereinafter termed "the primary RNA hairpin") consisting of the shRNA hairpin and at least one of the 5' single-stranded flanking RNA sequence and the 3' single- stranded flanking RNA sequence.
  • the shRNA design comprises the first flanking nucleotide sequence but not the second flanking nucleotide sequence.
  • the primary RNA hairpin consists of in 5' to 3' order the 5' single-stranded flanking RNA sequence and the shRNA hairpin.
  • the shRNA design comprises the second flanking nucleotide sequence but not the first flanking nucleotide sequence.
  • the primary RNA hairpin consists of in 5' to 3' order the shRNA hairpin and 3' single-stranded flanking RNA sequence.
  • the shRNA design comprises both the first flanking nucleotide sequence and the second flanking nucleotide sequence.
  • the primary RNA hairpin consists of in 5' to 3' order the 5' single-stranded flanking RNA sequence, the shRNA hairpin, and the 3' single-stranded flanking RNA sequence.
  • the first flanking nucleotide sequence preferably consists of at least 3 nucleotides.
  • the first flanking nucleotide sequence consists of at least 5, 8, 10, 15, 20, 30, 40 or 50 nucleotides.
  • the second flanking nucleotide sequence preferably consists of at least 3 nucleotides.
  • the second flanking nucleotide sequence consists of at least 5, 10, 15, 20, 30, 40 or 50 nucleotides.
  • the first and the second flanking nucleotide sequences are such that the encoded 5' single-stranded flanking RNA sequence and the 3' single-stranded flanking RNA sequence do not form any secondary structures under physiological conditions (e.g., at approximately 37 0 C and a pH level around 7.0).
  • flanking nucleotide sequences may be selected by selecting flanking nucleotide sequences that have been previously shown to discourage secondary structures formation during in vitro analysis where any secondary structures of the RNA molecules are monitored by spectroscopic methods such as UV absorbance.
  • the flanking nucleotide sequences are selected based on a computer-based prediction algorithm, e.g., original compiled based on experimental data.
  • the first and/or the second flanking nucleotide sequences is such that the encoded 5' single-stranded flanking RNA sequence and/or the 3' single-stranded flanking RNA sequence facilitates cleavage of the primary RNA hairpin that are produced from the shRNA design by Drosha.
  • the shRNA design comprises a Pol III promoter, e.g., a human or mouse Hl promoter or a human or mouse U6 promoter.
  • the primary RNA hairpin is produced by transcription by Pol III.
  • the first flanking nucleotide sequence starts with a guanine (G) at the 5' end.
  • G guanine
  • the invention provides a method utilizing Pol III for producing a primary RNA hairpin.
  • the DNA construct comprises a Pol II promoter. In such an embodiment, when the shRNA design is introduced into a cell expressing or containing Pol H, the primary RNA hairpin is produced by transcription by Pol II.
  • the promoter can be an inducible promoter such that the expression of the shRNA and hence the silencing of its target gene can be turned on or off as desired. Inducible expression of an siRNA is particularly useful for targeting essential genes.
  • the expression of the shRNA is under the control of a regulated promoter that allows tuning of the silencing level of the target gene.
  • a tetracycline regulated gene expression system is used (see, e.g., Gossen et al, 1995, Science 268:1766-1769; U.S. Patent No. 6,004,941; each of which is hereby incorporated herein by reference in its entirety).
  • an ecdysone-regulated gene expression system (see, e.g., Saez et al, 2000, Proc. Natl. Acad. ScL USA 97:14512-14517, which is hereby incorporated herein by reference in its entirety), or an MMTV glucocorticoid response element regulated gene expression system (see, e.g., Lucas et al., 1992, Annu. Rev. Biochem. 61:1131, which is hereby incorporated herein by reference in its entirety) may be used to regulate the expression of the shRNA.
  • the shRNA can be expressed from a plasmid (or virus).
  • the shRNA is expressed from recombinant vectors introduced either transiently or stably integrated into the genome of the target cells (see, e.g., Paddison et al, 2002, Genes Dev 16:948-958; Sui et al, 2002, Proc Natl Acad Sci USA 99:5515-5520; Yu et al, 2002, Proc Natl Acad Sci USA 99:6047-6052; Miyagishi et al, 2002, Nat Biotechnol 20:497-500; Paul et al, 2002, Nat Biotechnol 20:505-508; Kwak et al, 2003, J Pharmacol Sci 93:214-217; Brummelkamp et al, 2002, Science 296: 550-553; Boden et al , 2003, Nucleic Acids Res 31 :5033-5038; Kawasaki et al
  • the invention also provides a method of identifying off-target genes of an siRNA.
  • an "off-target" gene is a gene which is directly silenced by an siRNA that is designed to target another gene (see, International application No. PCT/US2004/015439 by Jackson et al., filed on May 17, 2004, which is incorporated herein by reference in its entirety).
  • An off-target gene can be silenced by either the sense strand or the antisense strand of the siRNA.
  • microarray experiments suggest that most siRNA oligos result in downregulation of off-target genes through direct interactions between an siRNA and the off-target transcripts. While sequence similarity between dsRNA and transcripts appears to play a role in determining which off-target genes are affected, sequence similarity searches, even combined with thermodynamic models of hybridization, are insufficient to predict off-target effects accurately. However, alignment of off-target transcripts with offending siRNA sequences reveals that some base pairing interactions between the two appear to be more important than others (Fig. 6).
  • the invention provides a method of identifying potential off-target genes of an siRNA using a PSSM that describes the sequence match pattern between an siRNA and a sequence of an off-target gene (pmPSSM).
  • Such a match pattern can be determined based on the frequency with which each position in an siRNA is found to match affected off-target transcripts identified as direct targets of the siRNA by simultaneous downregulation with the intended target through kinetic analysis of expression profiles (see, International application No.
  • An exemplary ⁇ P, ⁇ for a 19mer siRNA sequence is plotted in FIG. 7 and listed in Table I. Table I Weights of an exemplary pmPSSM for 21nt siRNAs having a 19 nt duplex region
  • sequence match pattern of off-target transcripts are used to obtain a pmPSSM.
  • Off-target genes of an siRNA can be identified using a method disclosed in International application No. PCT/US2004/015439 by Jackson et al., filed on May 17, 2004, which is hereby incorporated by reference herein in its entirety.
  • off-target genes of an siRNA are identified based on silencing kinetics (see, e.g., International application No. PCT/US2004/015439 by Jackson et al., filed on May 17, 2004, which is hereby incorporated by reference herein in its entirety).
  • a pmPSSM can then be generated using the frequency of matches found for each position.
  • the alignment shown in Fig. 6 and similar data for other siRNAs were combined to generate the exemplary position-specific scoring matrix for use in predicting off-target effects.
  • the degree of match between an siRNA and a sequence in a transcript can be evaluated with the pmPSSM using a score (also referred to as a position match score, pmScore) according to the following equation
  • L is the length of the alignment, e.g., 19.
  • a pmScore above a given threshold identifies the sequence as a potential off-target sequence.
  • the score threshold can be optimized by maximizing the correlation between predicted and observed numbers of off-target effects (Fig. 8).
  • the optimized threshold can be used to favor selection of siRNAs with relatively small numbers of predicted off-target effects.
  • Off-target genes of a given siRNA can be identified by first identifying off-target transcript sequences that align with the siRNA. Any suitable method for pair-wise alignment, such as but not limited to BLAST and FASTA, can be used. The position- specific scoring matrix is then used to calculate position match scores for these alignments. In a preferred embodiment, alignments are established with a low-stringency FASTA search and the score for each alignment is calculated according to Eq. 6. A score above a given threshold identifies the transcript comprising the sequence as a potential off-target gene.
  • the invention thus also provides a method of evaluating the silencing specificity of an siRNA.
  • potential off-target genes of the siRNA are identified.
  • the total number of such off-target genes in the genome or a portion of the genome is then used as a measure of the silencing specificity of the siRNA.
  • the invention provides a method for predicting strand preference and/or the efficacy and specificity of siRNAs based on position specific base composition of the siRNAs. It has been discovered that an siRNA whose base composition PSSM score (see Section 5.1.) is greater than the base composition PSSM (G/C PSSM) score of its reverse complement is predicted to have an antisense strand that is more active than its sense strand. In contrast, an siRNA whose base composition PSSM score is less than the base composition PSSM score of its reverse complement is predicted to have a sense strand that is more active than its antisense strand.
  • FIG. 14A shows the difference between the mean G/C content of the reverse complements of bad siRNAs with the mean G/C content of the bad siRNAs themselves, within the 19mer siRNA duplex region. The difference between the mean G/C content of good and bad siRNAs is shown for comparison. The curves are smoothed over a window of 5 (or portion of a window of 5, at the edges of the sequence).
  • FIG. 14B shows the difference between the mean G/C content of the reverse complements of good siRNAs with the mean G/C content of bad siRNAs, within the 19mer siRNA duplex region. The difference between the mean G/C content of good and bad siRNAs is shown for comparison. The curves are smoothed over a window of 5 (or portion of a window of 5, at the edges of the sequence).
  • siRNAs were binned by measured silencing efficacy, and the frequency of sense-active calls by the 3 '-biased method and G/C PSSM method was compared. Although these techniques are based on different analyses, the agreement is quite good. Both show that a higher proportion of low-silencing siRNAs vs. high-silencing siRNAs are predicted to be sense active.
  • the correlation coefficient for (siRNA G/C PSSM score - reverse complement G/C PSSM score) vs. logio(sense-identity score/antisense-identity score) is 0.59 for the set of 61 siRNAs binned in FIG. 15.
  • the invention provides a method for predicting strand preference, i.e., which of the two strands is move active, of siRNAs based on position specific base composition of the siRNAs.
  • the method comprises evaluating the strand preference of an siRNA in gene silencing by comparing the base compositions of the sense and the antisense strands of the siRNA.
  • the method comprises evaluating the strand preference of an siRNA in gene silencing by comparing the base compositions of the sense and the reverse complement of the target sequence of the siRNA.
  • the sequence of the antisense strand of an siRNA or the reverse complement of the target sequence of the siRNA in a transcript is compared with the target sequence using a PSSM approach (see Section 5.1.).
  • An siRNA and its reverse complement are scored using a PSSM based on a smoothed G/C content difference between good and bad siRNAs within the duplex region as the weight matrix.
  • a base composition weight matrix as described by FIG. 14A is used as the weight matrix.
  • the PSSM score of each strand can be calculated as the dot product of the siRNA strand G/C content with the G/C content difference matrix (as the score calculation method of curve model PSSMs).
  • an siRNA is identified as sense-active if its reverse complement PSSM score exceeded its own PSSM score.
  • the 3 '-biased method as described in International application Number PCT/US2004/015439 by Jackson et al., filed on May 17, 2004, which is incorporated by reference herein in its entirety, is used in conjunction with the PSSM score to determining the strand preference of an siRNA.
  • an siRNA is identified as sense-active by the 3 '-biased method of strand preference determination if the antisense-identical score exceeded the sense-identical score.
  • the invention also provides a method for identifying siRNAs having good silencing efficacy.
  • the method comprises identifying siRNAs having dominant antisense strand activity ("antisense-active" siRNAs) as siRNAs having good silencing efficacy and specificity (for silencing sense-identical target).
  • antisense-active siRNAs identifying siRNAs having dominant antisense strand activity
  • specificity for silencing sense-identical target.
  • the method described in Section 5.1 is used to identify siRNAs having good sense strand (i.e., identifying siRNAs having good silencing efficacy towards an antisense-identical target). Such siRNAs are then eliminated from uses in silencing sense-identical targets.
  • the method can also be used to eliminate siRNAs with dominant sense strand activity ("sense-active" siRNAs) as siRNAs having less efficacy and specificity for silencing sense-identical targets.
  • siRNAs with dominant sense strand activity (sense-active" siRNAs)
  • the reverse complements of bad siRNAs on the average, appear to have a GC content profile which differs from that of bad siRNAs in the same manner as the GC content profile of good siRNAs differs from that of bad siRNAs.
  • the reverse complements of bad siRNAs show even more extreme differences from bad siRNAs than do the good siRNAs. This observation is in accord with the evidence in siRNA expression profiles that many bad siRNAs have active sense strands.
  • the invention also provides a method for selecting siRNAs based on the base composition of the sequence of a reverse complement of the sense strand of the siRNAs.
  • a plurality of different siRNAs designed for silencing a target gene in an organism at a different target sequence in a transcript of the target gene is ranked according to positional base composition of the reverse complement sequences of their sense strands.
  • One or more siRNAs whose reverse complement sequences' positional base composition matches the positional base composition of desired siRNAs can then be selected.
  • the ranking of siRNAs is carried out by first determining a score for each different siRNA using a position-specific score matrix. The siRNAs are then ranked according to the score. Any method described in Section 5.1., supra, can be used to score reverse complement sequences.
  • the score for each reverse complement is calculated according to equation
  • the first type of siRNA can consist of one or more siRNAs having silencing efficacy no less than a first threshold, e.g., 75%, 80% or 90% at a suitable concentration, e.g., 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, preferably at 100 nM
  • the second type of siRNA can consist of one or more siRNAs having silencing efficacy less than a second threshold, e.g., 25%, 50%, or 75% at a suitable concentration, e.g., 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, preferably at 100 nM
  • the difference in probability is described by a sum of Gaussian curves, each of said Gaussian curves representing the difference in probability of finding a G or C at a different sequence position.
  • the methods of this invention can also be applied to developing models, e.g., PSSMs, of siRNA functional motifs by training position-specific scoring matrices to distinguish between bad siRNAs and their reverse complements (see, e.g., Section 5.1.).
  • a restriction in this analysis is that the reverse complements of bad siRNAs have no designated targets.
  • position-specific scoring matrices of 19mer siRNA duplex sequences are trained to distinguish between bad siRNAs and their reverse complements. Flanking sequence training can be performed on off-target genes in the case of distinguishing between bad siRNAs and their reverse complements, as well as in the case of distinguishing between any two groups of siRNAs.
  • the off-target activity of siRNAs can be hypothesized to have the same flanking sequence requirements as the on-target activity, as the same RNA-protein complexes are thought to be involved in both processes.
  • the regions flanking the alignment of the siRNA with the directly regulated off-target genes can be used to train and test models of flanking sequence requirements.
  • These models can be developed by any of the methods of this invention: random hill-climbing PSSMs, curve-model PSSMs, good-bad difference frequency matrices, good-composition frequency matrices, and/or bad-composition frequency matrices, etc.
  • the invention provides a method for designing siRNAs for gene silencing.
  • the method can be used to design siRNAs that have full sequence homology to their respective target sequences in a target gene.
  • the method can also be used to design siRNAs that have only partial sequence homology to a target gene.
  • Methods and compositions for silencing a target gene using an siRNA that has only partial sequence homology to its target sequence in a target gene is disclosed in International application No. PCT/US2004/015439 by Jackson et al., filed on May 17, 2004, which is incorporated by reference herein in its entirety.
  • an siRNA that comprises a sense strand contiguous nucleotide sequence of 11-18 nucleotides that is identical to a sequence of a transcript of the target gene but the siRNA does not have full length homology to any sequences in the transcript may be used to silence the transcript.
  • Such contiguous nucleotide sequence is preferably in the central region of the siRNA molecules.
  • a contiguous nucleotide sequence in the central region of an siRNA can be any continuous stretch of nucleotide sequence in the siRNA which does not begin at the 3' end.
  • a contiguous nucleotide sequence of 11 nucleotides can be the nucleotide sequence 2-12, 3-13, 4-14, 5-15, 6-16, 7-17, 8-18, or 9-19.
  • the contiguous nucleotide sequence is 11-16, 11-15, 14-15, 11, 12, or 13 nucleotides in length.
  • an siRNA that comprises a 3' sense strand contiguous nucleotide sequence of 9-18 nucleotides which is identical to a sequence of a transcript of the target gene but which siRNA does not have full length sequence identity to any contiguous sequences in the transcript may also be used to silence the transcript.
  • a 3' 9-18 nucleotide sequence is a continuous stretch of nucleotides that begins at the first paired base, i.e., it does not comprise the two base 3' overhang.
  • the contiguous nucleotide sequence is 9-16, 9-15, 9-12, 11, 10, or 9 nucleotides in length.
  • the method of Section 5.1 is used for identifying from among a plurality of siRNAs one or more siRNAs that have high silencing efficacy.
  • each siRNA in the plurality of siRNAs is evaluated for silencing efficacy by base composition PSSMs.
  • this step comprises calculating one or more PSSM scores for each siRNA.
  • the plurality of siRNAs is then ranked based on the scores and one or more siRNAs are selected using a method described in Section 5.1.4.
  • the method of Section 5.2 is used for identifying from among a plurality of siRNAs one or more siRNAs that have high silencing specificity.
  • alignments of each siRNA with sequences in each of a plurality of non-target transcripts are identified and evaluated with the pmPSSM approach (see Section 5.2.).
  • a pmScore is calculated for each of the alignments.
  • a pmScore above a given threshold identifies a sequence as a potential off-target sequence.
  • Such a pmScore is also termed an alignment score.
  • a pmScore can be a weighted FASTA alignment score.
  • the transcript that comprises the potential off-target sequence is identified as a potential off-target transcript.
  • the total number of such off-target transcripts in the genome or a portion of the genome is used as a measure of the silencing specificity of the siRNA.
  • One or more siRNAs having less off- target transcripts may then be selected.
  • sequence diversity is also referred to as "sequence variety” or simply “diversity” or “variety.” Sequence diversity can be represented or measured based on some sequence characteristics.
  • the siRNAs can be selected such that a plurality of siRNAs targeting a gene comprises siRNAs exhibiting sufficient difference in one or more of such diversity characteristics.
  • sequence diversity characteristics used in the method of the invention are quantifiable.
  • sequence diversity can be measured based on GC content, the location of the siRNA target sequence along the length of the target transcript, or the two bases upstream of the siRNA duplex (i.e., the leading dimer, with 16 different possible leading dimers).
  • the difference of two siRNAs can be measured as the difference between values of a sequence diversity measure.
  • the diversity or variety of a plurality of siRNAs can be quantitatively represented by the minimum difference or spacing in a sequence diversity measure between different siRNAs in the plurality.
  • the step of selection of siRNAs for diversity or variety is also referred to as a "de-overlap" step.
  • the de-overlapping selects siRNAs having differences of a sequence diversity measure between two siRNAs above a given threshold. For example, de-overlapping by position establishes a minimum distance between selected oligos along the length of the transcript sequence. In one embodiment, siRNAs positioned at least 100 bases apart in the transcript are selected. De-overlapping by GC content establishes a minimum difference in GC content. In one embodiment, the minimum difference in GC content is 1%, 2% or 5%.
  • De-overlapping by leading dimers establishes the probability of all or a portion of the 16 possible leading dimers among the selected siRNAs.
  • each of the 16 possible dimers is assigned a score of 1-16, and a 0.5 is used to select all possible leading primer with equal probability.
  • the candidates are preferably de-overlapped on GC content, with a minimum spacing of 5%, a maximum number of duplicates of each value of GC% of 100 and at least 200 candidates selected; more preferably they are de-overlapped on GC content with a minimum spacing of 5%, a maximum number of duplicates of each value of GC% of 80 and at least 200 candidates selected; and still more preferably they are de- overlapped on GC content with a minimum spacing of 5%, a maximum number of duplicates of each value of GC% of 60 and at least 200 candidates selected.
  • siRNAs can be further selected based additional selection criteria.
  • siRNAs targeting sequences not common to all documented splice forms are eliminated.
  • siRNAs targeting sequences overlapping with simple or interspersed repeat elements are eliminated.
  • siRNAs targeting sequences positioned at least 75 bases downstream of the translation start codon are selected.
  • siRNAs targeting sequences overlapping or downstream of the stop codon are eliminated. This avoids targeting sequences absent in undocumented alternative polyadenylation forms.
  • siRNAs with GC content close to 50% are selected.
  • siRNAs with GC% ⁇ 20% and > 70% are eliminated.
  • 10% ⁇ GC% ⁇ 90%, 20% ⁇ GC% ⁇ 80%, 25% ⁇ GC% ⁇ 75%, 30% ⁇ GC% ⁇ 70% are retained.
  • siRNAs targeting sequence containing 4 consecutive guanosine, cytosine, adenine or uracil residues are eliminated.
  • siRNAs targeting a sequence with a guanine or cytosine residue at the first position in the 19mer duplex region at the 5' end are selected. Such siRNAs target sequences that are effectively transcribed by RNA polymerase III.
  • siRNAs targeting a sequence containing recognition sites for one or more given restriction endonucleases, e.g., Xhol or EcoRI restriction endonucleases are eliminated. This embodiment may be used to select siRNAs sequences for construction of the shRNA vectors.
  • the siRNAs are evaluated for binding energy. See WO 01/05935 for an exemplary method of determining binding energy.
  • the binding energy is evaluated by calculating the nearest-neighbor 21mer ⁇ G.
  • the siRNAs are evaluated for binding specificity. See WO 01/05935 for an exemplary method of determining binding specificity of a 21mer.
  • the method for predicting strand preference and/or the efficacy and specificity of siRNAs based on position specific base composition of the siRNAs as described in Section 5.3 can be used to evaluate the siRNA candidates.
  • FIG. 9 A flow chart of an exemplary embodiment of the method used to select the siRNAs is shown in FIG. 9.
  • siRNA sequences that target a transcript are selected. In one embodiment, all 19mer subsequences of the transcript are considered. The appropriate flanking sequences for each siRNA sequence are also obtained and considered.
  • the siRNAs are evaluated against the following filters: (1) eliminating siRNAs targeting sequences not common to all documented splice forms; (2) eliminating siRNAs targeting sequences overlapping with simple or interspersed repeat elements; (3) eliminating siRNAs targeting sequences positioned within 75 bases downstream of the translation start codon; and (4) eliminating siRNAs overlapping or downstream of the stop codon.
  • siRNAs targeting sequence containing 4 consecutive guanosine, cytosine, adenine or uracil residues are also taken: (5) eliminating siRNAs targeting sequence containing 4 consecutive guanosine, cytosine, adenine or uracil residues; (6) retaining siRNAs targeting a sequence with a guanine residue at the first and/or the second position in the 19mer duplex region at the 5' end; and (7) eliminating siRNAs targeting a sequence containing recognition sites for one or more given restriction enzymes, e.g., Xhol or EcoRI restriction endonucleases, if siRNAs sequences used in construction of the shRNA vectors.
  • restriction enzymes e.g., Xhol or EcoRI restriction endonucleases
  • step 102 the siRNA is evaluated for silencing efficacy by base composition PSSMs.
  • step 102 comprises calculating a first PSSM score, i.e., the PSSM-I score, and a second PSSM score, i.e., the PSSM-2 score, for an siRNA.
  • the two scores are sum to calculate the combined PSSM- l+PSSM-2 score for the siRNA.
  • the PSSMs used are those whose performance is shown in Figure 2.
  • the siRNA is retained if the combined score is above a given threshold.
  • the siRNA is then evaluated for its binding energy by calculating the nearest- neighbor 21mer ⁇ G.
  • the siRNA is then evaluated for binding specificity by calculating a 21mer minimax score against the set of unique sequence representatives of genes of an organism, e.g., the set of unique sequences representatives for each cluster of Homo sapiens Unigene build 161. See WO 01/05935 for methods of calculating the ⁇ G and the minimax score.
  • the parameters for the BLAST alignments and nearest-neighbor delta-G calculations based on the BLAST alignments, which are used to compute minimax scores are as follows: -p blastn -e 100 -F F -W 11 -b 200 -v 10000 -S 3; and delta-G: temperature 66°; salt IM; concentration 1 pM; type of nucleic acid, RNA.
  • the siRNA is eliminated if the (21mer ⁇ G - 21mer minimax) ⁇ 0.5.
  • siRNAs are screened for overall GC content. In one embodiment, siRNAs with GC content significantly deviated from 50%, e.g., GC% ⁇ 20% and > 70%, are eliminated.
  • siRNAs are screened for diversity or variety. Position simply refers to the position of the oligo in the transcript sequence and is automatically provided by identifying the oligo. Variety is enforced in one or more "de-overlap" steps in the method. Briefly, de-overlapping selects for above-threshold spacing between selected oligos in some calculable parameter. To de-overlap, oligos are first ranked according to some parameter thought to distinguish better from poorer performers and then selected for spacing between oligos according to some other parameter. To begin, the top ranked oligo is selected. Then the ranked list is examined, and the next-best oligo with at least the minimum required spacing from the selected oligo is selected.
  • next-best oligo with at least the minimum spacing from the two selected oligos is also selected.
  • the process continues until the desired number of oligos is selected.
  • multiple oligos may share the same value if a parameter is few- valued, and the number of oligos sharing the same value is limited by a set threshold.
  • the spacing requirement can be relaxed until the desired number, or the set of all remaining available oligos, is selected.
  • de-overlapping by position establishes a minimum distance between selected oligos along the length of the transcript sequence.
  • siRNAs are ranked by a PSSM score and the ranked siRNAs positioned at least 100 bases apart in the transcript are selected.
  • De-overlapping by GC content establishes a minimum difference in GC content, hi one embodiment, the minimum difference in GC content is 1%, 2% or 5%. Duplicates are allowed for few- valued parameters such as the GC% of a 19mer.
  • De-overlapping by leading dimers establishes the probability of all or a portion of the 16 possible leading dimers among the selected siRNAs. In one embodiment, each of the 16 possible dimers is assigned a score of 1-16, and a 0.5 is used to selected all possible leading primer with equal probability, i.e., to distribute candidate siRNAs over all possible leading dimer values.
  • step 105 off-target activity of an siRNA is evaluated according to the method described in Section 5.2. Alignments of each siRNA with sequences in each of a plurality of non-target transcripts are identified and evaluated with a pmPSSM using a pmScore calculated according to equation (6). A pmScore above a given threshold identifies the sequence as a potential off-target sequence. The transcript that comprises the potential off-target sequence is identified as a potential off-target transcript. The total number of such off-target transcripts in the genome or a portion of the genome is used as a measure of the silencing specificity of the siRNA. One or more siRNAs having less off-target transcripts are selected.
  • transcripts of genes are scanned using FASTA with the parameters: KTUP 6 -r 3/-7 -g -6 -f -6 -d 14000 -b 14000 -E 7000.
  • a pmScore is determined for each alignment as described in Section 5.2.
  • the FASTA weighted score is used to: (1) quantify the nearest sequence match to the candidate siRNA; and (2) count the total matches to the candidate siRNA with weighted scores over a threshold. The total number of such off-target genes in the genome or a portion of the genome is then used as a measure of the silencing specificity of the siRNA.
  • the selected siRNAs are subjected to a second round of selection for variety (step 106), and re-ranked by their base composition PSSM scores (step 107). The desired number of siRNAs is retained from the top of this final ranking (step 108).
  • the invention also provides a method for selecting a plurality of siRNAs for each of a plurality of different genes, each siRNA achieving at least 75%, at least 80%, or at least 90% silencing of its target gene.
  • the method described above is used to select a plurality of siRNAs for each of a plurality of genes.
  • the plurality of siRNAs consists of at least 3, 5, or 10 siRNAs.
  • the plurality of different genes consists of at least 100, 500, 1,000, 5,000, 10,000 or 30,000 different genes.
  • the invention also provides a library of siRNAs which comprises a plurality of siRNAs for each of a plurality of different genes, each siRNA achieves at least 75%, at least 80%, or at least 90% silencing of its target gene.
  • the standard conditions are 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, siRNA, preferably at 100 nM, silencing assayed by TaqMan 24 hours post-transfection.
  • the plurality of siRNAs consists of at least 3, at least 5, or at least 10 siRNAs.
  • the plurality of different genes consists of at least 10, 100, 500, 1,000, 5,000, 10,000 or 30,000 different genes.
  • any standard method for gene silencing can be used in conjunction with the present invention, e.g., to carry our gene silencing using siRNAs designed by a method described in the present invention (see, e.g. , Guo et al , 1995, Cell 81 :611 -620; Fire et al.
  • gene silencing is induced by presenting the cell with the siRNA, mimicking the product of Dicer cleavage (see, e.g., Elbashir et al, 2001, Nature 411:494-498; and Elbashir et al, 2001, Genes Dev. 15:188- 200, each of which is incorporated by reference herein in its entirety).
  • Synthetic siRNA duplexes maintain the ability to associate with RISC and direct silencing of mRNA transcripts.
  • siRNAs can be chemically synthesized, or derived from cleavage of double- stranded RNA by recombinant Dicer. Cells can be transfected with the siRNA using standard method known in the art.
  • siRNA transfection is carried out as follows: one day prior to transfection, 100 microliters of chosen cells, e.g., cervical cancer HeLa cells (ATCC, Cat. No. CCL-2), grown in DMEM/10% fetal bovine serum (Invitrogen, Carlsbad, CA) to approximately 90% confluency are seeded in a 96- well tissue culture plate (Corning, Corning, NY ) at 1500 cells/well. For each transfection 85 microliters of OptiMEM (Invitrogen) is mixed with 5 microliter of serially diluted siRNA (Dharmacon, Denver) from a 20 micro molar stock.
  • chosen cells e.g., cervical cancer HeLa cells (ATCC, Cat. No. CCL-2)
  • DMEM/10% fetal bovine serum Invitrogen, Carlsbad, CA
  • OptiMEM Invitrogen
  • siRNA serially diluted siRNA
  • OptiMEM For each transfection 5 microliter OptiMEM is mixed with 5 microliter Oligofectamine reagent (Invitrogen) and incubated 5 minutes at room temperature. The 10 microliter OptiMEM/Oligofectamine mixture is dispensed into each tube with the OptiMEM/siRNA mixture, mixed and incubated 15-20 minutes at room temperature. 10 microliter of the transfection mixture is aliquoted into each well of the 96-well plate and incubated for 4 hours at 37°C and 5% CO 2 .
  • RNA interference is carried out using pool of siRNAs.
  • the total siRNA concentration of the pool is about the same as the concentration of a single siRNA when used individually, e.g., 10 ⁇ M, l ⁇ M, 100 nM, 50 nM, 10 nM or 1 nM, preferably at 100 nM.
  • the total concentration of the pool of siRNAs is an optimal concentration for silencing the intended target gene.
  • An optimal concentration is a concentration further increase of which does not increase the level of silencing substantially.
  • the optimal concentration is a concentration further increase of which does not increase the level of silencing by more than 5%, 10% or 20%.
  • the composition of the pool including the number of different siRNAs in the pool and the concentration of each different siRNA, is chosen such that the pool of siRNAs causes less than 30%, 20%, 10% or 5%, 1%, 0.1% or 0.01% of silencing of any off-target genes.
  • the concentration of each different siRNA in the pool of different siRNAs is about the same.
  • the respective concentrations of different siRNAs in the pool are different from each other by less than 5%, 10%, 20% or 50%.
  • at least one siRNA in the pool of different siRNAs constitutes more than 90%, 80%, 70%, 50%, or 20% of the total siRNA concentration in the pool.
  • none of the siRNAs in the pool of different siRNAs constitutes more than 90%, 80%, 70%, 50%, or 20% of the total siRNA concentration in the pool.
  • each siRNA in the pool has an concentration that is lower than the optimal concentration when used individually.
  • each different siRNA in the pool has an concentration that is lower than the concentration of the siRNA that is effective to achieve at least 30%, 50%, 75%, 80%, 85%, 90% or 95 % silencing when used in the absence of other siRNAs or in the absence of other siRNAs designed to silence the gene.
  • each different siRNA in the pool has a concentration that causes less than 30%, 20%, 10% or 5% of silencing of the gene when used in the absence of other siRNAs or in the absence of other siRNAs designed to silence the gene.
  • each siRNA has a concentration that causes less than 30%, 20%, 10% or 5% of silencing of the target gene when used alone, while the plurality of siRNAs causes at least 80% or 90% of silencing of the target gene.
  • Another method for gene silencing is to introduce into a cell an shRNA, for short hairpin RNA (see, e.g., Paddison et al, 2002, Genes Dev. 16:948-958; Brummelkamp et al, 2002, Science 296:550-553; and Sui, G. et al, 2002, Proc. Natl. Acad. Sci. USA 99:5515-5520, each of which is incorporated by reference herein in its entirety), which can be processed in the cells into siRNA.
  • a desired siRNA sequence is expressed from a plasmid (or virus) as an inverted repeat with an intervening loop sequence to form a hairpin structure.
  • RNA transcript containing the hairpin is subsequently processed by Dicer to produce siRNAs for silencing.
  • Plasmid-based shRNAs can be expressed stably in cells, allowing long-term gene silencing in cells both in vitro and in vivo, e.g., in animals (see, McCaffrey et al., 2002, Nature 418:38-39; Xia et al., 2002, Nat. Biotech. 20:1006-1010; Lewis et al, 2002, Nat. Genetics 32:107-108; Rubinson et al, 2003, Nat: Genetics 33:401-406; Tiscornia et al, 2003, Proc. Natl. Acad. Sci. USA 100:1844-1848, each of which is incorporated by reference herein in its entirety).
  • a plasmid-based shRNA is used.
  • shRNAs are expressed from recombinant vectors introduced either transiently or stably integrated into the genome (see, e.g., Paddison et al, 2002, Genes Dev 16:948-958; Sui et al, 2002, Proc Natl Acad Sci USA 99:5515-5520; Yu et al, 2002, Proc Natl Acad Sci USA 99:6047-6052; Miyagishi et al, 2002, Nat Biotechnol 20:497-500; Paul et al, 2002, Nat Biotechnol 20:505-508; Kwak et al, 2003, J Pharmacol Sci 93:214-217; Brummelkamp et al, 2002, Science 296:550-553; Boden et al , 2003, Nucleic Acids Res 31 :5033-5038; Kawasaki et al, 2003, Nucleic Acids Res 31 :700-707; each of which is incorporated by reference herein in its
  • the siRNA that disrupts the target gene can be expressed (via an shRNA) by any suitable vector which encodes the shRNA.
  • the vector can also encode a marker which can be used for selecting clones in which the vector or a sufficient portion thereof is integrated in the host genome such that the shRNA is expressed. Any standard method known in the art can be used to deliver the vector into the cells.
  • cells expressing the shRNA are generated by transfecting suitable cells with a plasmid containing the vector. Cells can then be selected by the appropriate marker. Clones are then picked, and tested for knockdown.
  • a plurality of recombinant vectors is introduced into the genome such that the expression level of the siRNA can be above a given value. Such an embodiment is particular useful for silencing genes whose transcript level is low in the cell.
  • the expression of the shRNA is under the control of an inducible promoter such that the silencing of its target gene can be turned on when desired. Inducible expression of an siRNA is particularly useful for targeting essential genes.
  • the expression of the shRNA is under the control of a regulated promoter that allows tuning of the silencing level of the target gene. This allows screening against cells in which the target gene is partially knocked out.
  • a regulated promoter refers to a promoter that can be activated when an appropriate inducing agent is present.
  • An “inducing agent” can be any molecule that can be used to activate transcription by activating the regulated promoter.
  • An inducing agent can be, but is not limited to, a peptide or polypeptide, a hormone, or an organic small molecule.
  • An analogue of an inducing agent i.e., a molecule that activates the regulated promoter as the inducing agent does, can also be used.
  • the level of activity of the regulated promoter induced by different analogues may be different, thus allowing more flexibility in tuning the activity level of the regulated promoter.
  • the regulated promoter in the vector can be any mammalian transcription regulation system known in the art (see, e.g., Gossen et al,
  • the regulated promoter is regulated in a dosage and/or analogue dependent manner.
  • the level of activity of the regulated promoter is tuned to a desired level by a method comprising adjusting the concentration of the inducing agent to which the regulated promoter is responsive.
  • the desired level of activity of the regulated promoter as obtained by applying a particular concentration of the inducing agent, can be determined based on the desired silencing level of the target gene.
  • a tetracycline regulated gene expression system is used (see, e.g., Gossen et al., 1995, Science 268:1766-1769; U.S. Patent No. 6,004,941 ; each of which is incorporated by reference herein in its entirety).
  • a tet regulated system utilizes components of the tet repressor/operator/inducer system of prokaryotes to regulate gene expression in eukaryotic cells.
  • the invention provides methods for using the tet regulatory system for regulating the expression of an shRNA linked to one or more tet operator sequences. The methods involve introducing into a cell a vector encoding a fusion protein that activates transcription.
  • the fusion protein comprises a first polypeptide that binds to a tet operator sequence in the presence of tetracycline or a tetracycline analogue operatively linked to a second polypeptide that activates transcription in cells.
  • a tetracycline or a tetracycline analogue
  • expression of the tet operator-linked shRNA is regulated.
  • an ecdyson regulated gene expression system see, e.g.,
  • the pRETRO- SUPER (pRS) vector which encodes a puromycin-resistance marker and drives shRNA expression from an Hl (RNA Pol III) promoter is used.
  • the pRS-shRNA plasmid can be generated by any standard method known in the art.
  • the pRS-shRNA is deconvoluted from the library plasmid pool for a chosen gene by transforming bacteria with the pool and looking for clones containing only the plasmid of interest.
  • a 19mer siRNA sequence is used along with suitable forward and reverse primers for sequence specific PCR. Plasmids are identified by sequence specific PCR, and confirmed by sequencing.
  • Cells expressing the shRNA are generated by transfecting suitable cells with the pRS-shRNA plasmid. Cells are selected by the appropriate marker, e.g., puromycin, and maintained until colonies are evident. Clones are then picked, and tested for knockdown.
  • an shRNA is expressed by a plasmid, e.g., a pRS-shRNA. The knockdown by the pRS-shRNA plasmid can be achieved by transfecting cells using Lipofectamine 2000 (Invitrogen).
  • siRNAs can be delivered to an organ or tissue in an animal, such a human, in vivo (see, e.g., Song et al, 2003, Nat. Medicine 9:347-351; Sorensen et al, 2003, J MoI Biol. 327:761-766; and Lewis et al, 2002, Nat. Genetics 32:107-108, each of which is incorporated by reference herein in their entirety).
  • a solution of siRNA is injected intravenously into the animal.
  • the siRNA can then reach an organ or tissue of interest and effectively reduce the expression of the target gene in the organ or tissue of the animal.
  • the siRNAs can also be delivered to an organ or tissue using a gene therapy approach. Any of the methods for gene therapy available in the art can be used to deliver the siRNA. For general reviews of the methods of gene therapy, see Goldspiel et al,
  • the therapeutic comprises a nucleic acid encoding the siRNA as a part of an expression vector.
  • such a nucleic acid has a promoter operably linked to the siRNA coding region, in which the promoter being inducible or constitutive, and, optionally, tissue-specific.
  • a nucleic acid molecule in which the siRNA coding sequence is flanked by regions that promote homologous recombination at a desired site in the genome is used (see e.g., Koller and Smithies, 1989, Proc. Natl. Acad. ScL U.S.A. 86:8932-8935; Zijlstra et al, 1989, Nature 342:435-438; each of which is incorporated by reference herein in its entirety).
  • the nucleic acid is directly administered in vivo.
  • nucleic acid-ligand complex can be formed in which the ligand comprises a fusogenic viral peptide to disrupt endosomes, allowing the nucleic acid to avoid lysosomal degradation.
  • the nucleic acid can be targeted in vivo for cell specific uptake and expression, by targeting a specific receptor (see, e.g., PCT Publications WO 92/06180 dated April 16, 1992 (Wu et al); WO 92/22635 dated December 23, 1992 (Wilson et al); WO92/20316 dated November 26, 1992 (Findeis et al); WO93/14188 dated July 22, 1993 (Clarke et al), WO 93/20221 dated October 14, 1993 (Young)).
  • the nucleic acid can be introduced intracellularly and incorporated within host cell DNA for expression, by homologous recombination (Koller and Smithies, 1989, Proc.
  • a viral vector that contains the siRNA coding nucleic acid is used.
  • a retroviral vector can be used (see Miller et al, 1993, Meth. Enzymol. 217: 581-599, which is incorporated by reference herein in its entirety). These retroviral vectors have been modified to delete retroviral sequences that are not necessary for packaging of the viral genome and integration into host cell DNA.
  • the siRNA coding nucleic acid to be used in gene therapy is cloned into the vector, which facilitates delivery of the gene into a patient. More detail about retroviral vectors can be found in Boesen et al., 1994, Biotherapy 6:291-302, which describes the use of a retroviral vector to deliver the mdrl gene to hematopoietic stem cells in order to make the stem cells more resistant to chemotherapy. Other references illustrating the use of retroviral vectors in gene therapy are: Clowes et al, 1994, J. Clin. Invest.
  • Adenoviruses are other viral vectors that can be used in gene therapy.
  • Adenoviruses are especially attractive vehicles for delivering genes to respiratory epithelia. Adenoviruses naturally infect respiratory epithelia where they cause a mild disease. Other targets for adenovirus-based delivery systems are liver, the central nervous system, endothelial cells, and muscle. Adenoviruses have the advantage of being capable of infecting non-dividing cells. Kozarsky and Wilson (1993, Current Opinion in Genetics and Development 3:499-503, which is incorporated by reference herein in its entirety) present a review of adenovirus-based gene therapy. Bout et al. (1994, Human Gene Therapy 5:3-1010) demonstrated the use of adenovirus vectors to transfer genes to the respiratory epithelia of rhesus monkeys.
  • Adeno-associated virus may also been used in gene therapy (Walsh et al., 1993, Proc. Soc. Exp. Biol. Med. 204:289-300, which is incorporated by reference herein in its entirety).
  • Degree of silencing can be determined using any standard RNA or protein quantification method known in the art.
  • RNA quantification can be performed using Real-time PCR, e.g., using AP Biosystems TaqMan pre-developed assay reagent (#4319442).
  • Primer probe for the appropriate gene can be designed using any standard method known in the art, e.g. using Primer Express software.
  • RNA values can be normalized to RNA for actin (#4326315).
  • Protein levels can be quantified by flow cytometry following staining with appropriate antibody and labeled secondary antibody. Protein levels can also be quantified by western blot of cell lysates with appropriate monoclonal antibodies followed by Kodak image analysis of chemiluminescent immunoblot. Protein levels can also be normalized to actin levels.
  • Effects of gene silencing on a cell can be evaluated by any known assay.
  • cell growth can be assayed using any suitable proliferation or growth inhibition assays known in the art.
  • an MTT proliferation assay see, e.g., van de Loosdrechet, et al., 1994, J. Immunol Methods 174:31 1-320; Ohno et al, 1991 , J. Immunol. Methods 145:199-203; Ferrari et al, 199O 5 J. Immunol Methods 131 : 165-172; Alley et al, 1988, Cancer Res. 48: 589-601; Carmichael et al, 1987, Cancer Res.
  • the cells are then incubated with a suitable amount of 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) for a chosen period of time, e.g., 1-8 hours, such that viable cells convert MTT into an intracellular deposit of insoluble formazan.
  • a suitable MTT solvent e.g., a DMSO solution
  • the concentration of MTT is then measured by determining the optical density at e.g. , 570 nm.
  • a plurality of different concentrations of the candidate agent can be assayed to allow the determination of the concentrations of the candidate agent or agents which causes 50% inhibition.
  • an alamarBlueTM Assay for cell proliferation is used to screen for one or more candidate agents that can be used to inhibit the growth of cells (see, e.g., Page et al., 1993, Int. J. Oncol. 3:473-476, which is incorporated by reference herein in its entirety).
  • An alamarBlueTM assay measures cellular respiration and uses it as a measure of the number of living cells. The internal environment of proliferating cells is more reduced than that of non-proliferating cells.
  • the ratios of NADPH/NADP, FADH/FAD, FMNH/FMN, and NADH/NAF increase during proliferation.
  • AlamarBlue can be reduced by these metabolic intermediates and, therefore, can be used to monitor cell proliferation.
  • the cell number of a treated sample as measured by alamarBlue can be expressed in percent relative to that of an untreated control sample.
  • alamarBlue reduction can be measured by either absorption or fluorescence spectroscopy. In one embodiment, the alamarBlue reduction is determined by absorbance and calculated as percent reduced using the equation:
  • (A' ⁇ 2 ) Absorbance of negative control wells which contain medium plus alamar Blue but to which no cells have been added at 600 nm.
  • the % Reduced of wells containing no cell was subtracted from the % Reduced of wells containing samples to determine the % Reduced above background.
  • Cell cycle analysis can be carried out using standard method known in the art.
  • the supernatant from each well is combined with the cells that have been harvested by trypsinization.
  • the mixture is then centrifuged at a suitable speed.
  • the cells are then fixed with, e.g., ice cold 70% ethanol for a suitable period of time, e.g., ⁇ 30 minutes.
  • the Sub-Gl cell population is used as a measure of cell death.
  • the cells are said to have been sensitized to an agent if the Sub-Gl population from the sample treated with the agent is larger than the Sub-Gl population of sample not treated with the agent.
  • Any of the methods of the present invention can preferably be implemented using an apparatus such as a computer system (of one or more computers), such as the computer system described in this section, according to the following programs and methods.
  • a computer system can also preferably store and manipulate measured signals obtained in various experiments that can be used by a computer system implemented with the analytical methods of this invention. Accordingly, such computer systems are also considered part of the present invention.
  • Computer system 1201 is illustrated here as comprising internal components and as being linked to external components.
  • the internal components of this computer system include one or more processor elements 1202 interconnected with a main memory 1203.
  • processor elements 1202 interconnected with a main memory 1203.
  • main memory 1203. can be an Intel Pentium IV®-based processor of 2 GHZ or greater clock rate and with 256 MB or more main memory.
  • computer system 1201 is a cluster of a plurality of computers comprising a head "node” and eight sibling "nodes,” with each node having a central processing unit (“CPU").
  • CPU central processing unit
  • the cluster also comprises at least 128 MB of random access memory ("RAM") on the head node and at least 256 MB of RAM on each of the eight sibling nodes. Therefore, the computer systems of the present invention are not limited to those consisting of a single memory unit or a single processor unit.
  • RAM random access memory
  • the external components can include a mass storage 1204.
  • This mass storage can be one or more hard disks that are typically packaged together with the processor and memory. Such hard disks are typically of 10 GB or greater storage capacity and more preferably have at least 40 GB of storage capacity.
  • each node can have its own hard drive.
  • the head node preferably has a hard drive with at least 10 GB of storage capacity whereas each sibling node preferably has a hard drive with at least 40 GB of storage capacity.
  • a computer system of the invention can further comprise other mass storage units including, for example, one or more floppy drives, one more CD-ROM drives, one or more DVD drives or one or more DAT drives.
  • a user interface device 1205 which is most typically a monitor and a keyboard together with a graphical input device 1206 such as a "mouse.”
  • the computer system is also typically linked to a network link 1207 which can be, e.g., part of a local area network (“LAN”) to other, local computer systems and/or part of a wide area network (“WAN”), such as the Internet, that is connected to other, remote computer systems.
  • LAN local area network
  • WAN wide area network
  • each node is preferably connected to a network, preferably an NFS network, so that the nodes of the computer system communicate with each other and, optionally, with other computer systems by means of the network and can thereby share data and processing tasks with one another.
  • a network preferably an NFS network
  • the software components comprise both software components that are standard in the art and components that are special to the present invention. These software components are typically stored on mass storage such as the hard drive 1204, but can be stored on other computer readable media as well including, for example, one or more floppy disks, one or more CD-ROMs, one or more DVDs or one or more DATs.
  • Software component 1210 represents an operating system which is responsible for managing the computer system and its network interconnections. The operating system can be, for example, of the Microsoft WindowsTM family such as Windows 95, Window 98, Windows NT, Windows 2000 or Windows XP.
  • the operating software can be a Macintosh operating system, a UNIX operating system or a LINUX operating system.
  • Software components 1211 comprises common languages and functions that are preferably present in the system to assist programs implementing methods specific to the present invention. Languages that can be used to program the analytic methods of the invention include, for example, C and C++, FORTRAN, PERL, HTML, JAVA, and any of the UNIX or LINUX shell command languages such as C shell script language.
  • the methods of the invention can also be programmed or modeled in mathematical software packages that allow symbolic entry of equations and high-level specification of processing, including specific algorithms to be used, thereby freeing a user of the need to procedurally program individual equations and algorithms. Such packages include, e.g., Matlab from Mathworks (Natick, MA), Mathematica from Wolfram Research (Champaign, IL) or S-Plus from MathSoft (Seattle, WA).
  • Software component 1212 comprises any analytic methods of the present invention described supra, preferably programmed in a procedural language or symbolic package.
  • software component 1212 preferably includes programs that cause the processor to implement steps of accepting a plurality of measured signals and storing the measured signals in the memory.
  • the computer system can accept measured signals that are manually entered by a user (e.g., by means of the user interface). More preferably, however, the programs cause the computer system to retrieve measured signals from a database.
  • a database can be stored on a mass storage (e.g., a hard drive) or other computer readable medium and loaded into the memory of the computer, or the compendium can be accessed by the computer system by means of the network 1207.
  • the present invention can be implemented as a computer program product that comprises a computer program mechanism embedded in a computer readable storage medium. Further, any of the methods of the present invention that don't involve a measuring step can be implemented in one or more computers or computer systems. Further still, any of the methods of the present invention that don't involve a measuring step can be implemented in one or more computer program products. Some embodiments of the present invention provide a computer system or a computer program product that encodes or has instructions for performing any or all of the methods disclosed herein. Such methods/instructions can be stored on a CD-ROM, DVD, magnetic disk storage product, or any other computer readable data or program storage product.
  • Such methods can also be embedded in permanent storage, such as ROM, one or more programmable chips, or one or more application specific integrated circuits (ASICs).
  • permanent storage can be localized in a server, 802.11 access point, 802.11 wireless bridge/station, repeater, router, mobile phone, or other electronic devices.
  • Such methods encoded in the computer program product can also be distributed electronically, via the Internet or otherwise, by transmission of a computer data signal (in which the software modules are embedded) either digitally or on a carrier wave.
  • Some embodiments of the present invention provide a computer program product that contains any or all of the program modules shown in Fig. 12. These program modules can be stored on a CD-ROM, DVD, magnetic disk storage product, or any other computer readable data or program storage product.
  • the program modules can also be embedded in permanent storage, such as ROM, one or more programmable chips, or one or more application specific integrated circuits (ASICs). Such permanent storage can be localized in a server, 802.11 access point, 802.11 wireless bridge/station, repeater, router, mobile phone, or other electronic devices.
  • the software modules in the computer program product can also be distributed electronically, via the Internet or otherwise, by transmission of a computer data signal (in which the software modules are embedded) either digitally or on a carrier wave.
  • the computer program products contain programs with instructions for carrying out all or part of any of the methods of the present invention.
  • the systems and methods of the present invention further comprises displaying or outputting to a user, to an output medium such as paper or a computer screen, to a computer readable storage medium, or to a remote computer, a result or an indicia of the present invention.
  • siRNAs targeting more than 700 genes were constructed.
  • the siRNAs in the library were designed by use of a "standard” approach, based on a combination of limited design principles available from the scientific literature (Elbashir et al., 2001, Nature 411 :494-8) and a method for predicting off target effects by sequence similarity scoring as described in Section 5.2.
  • a set of 377 siRNAs was tested by Taqman analysis for their ability to silence their respective target genes. The set of 377 siRNAs are listed in Table II.
  • Table II lists the following information for the 377 siRNAs: the ID number of the siRNA, the accession number of the target gene, start position of the target sequence, target sequence, % silencing, the set it belongs (i.e., training or test) in Set 1, the set it belongs in Set 2, and the SEQ ID NO.
  • the results of this test showed that most siRNAs successfully silenced their target genes (median silencing, -75%), but individual siRNAs still showed a wide range of silencing performance. Good (or poor) silencing ability was not consistently associated with any particular base at any position, overall GC content, the position of the siRNA sequence within the target transcript, or with alternative splicing of target transcripts.
  • siRNAs were divided into groups containing those with less than median silencing ability ("bad" siRNAs) and those with median or better silencing ability ("good” siRNAs).
  • a number of metrics were evaluated for their ability to distinguish good and bad siRNAs, including base composition in windows of the 19mer siRNA duplex sequence and the flanking target region, secondary structure predictions by various programs and thermodynamic properties.
  • the GC content of good siRNAs differed substantially from that of bad siRNAs in a position- specific manner (FIGS. 1-3).
  • good siRNA duplexes were not observed to be associated with any particular sequence, but tended to be GC rich at the 5' end and GC poor at the 3' end.
  • the data indicate that a good siRNA duplex encourages preferential interaction of the antisense strand by being GC poor at its 3' end and discourages interaction of the sense strand by being GC rich at its 5' end.
  • position-specific sequence preferences extend beyond the boundaries of the siRNA target sequence into the adjacent sequence(s). This suggests that steps during RNA silencing other than unwinding of the siRNA duplex are affected by position-specific base composition preferences.
  • the GC-content difference between good and bad siRNAs shown in FIGS. 1 and 2 was used to develop methods for selecting good siRNAs. Best results were obtained with a position-specific scoring matrix (PSSM) approach.
  • PSSM position-specific scoring matrix
  • the PSSM provides weights for GC, A or U at every position on the sense strand of the target gene sequence from 10 bases upstream of the start to 10 bases downstream of the end of the siRNA duplex.
  • the siRNA efficacy data were divided into two sets, one to be used for training and the other for an independent test.
  • a random-mutation hill-climbing search algorithm was used to optimize the weights for each base at each position of the PSSM simultaneously.
  • the optimization criterion was the correlation coefficient between the target silencing of the siRNA and its PSSM score. Multiple runs of optimization on the training data set were averaged to complete each PSSM. Each PSSM was then tested on the independent (test) set of siRNAs. The performance of two PSSMs on their training and test data sets is demonstrated in Figure
  • siRNA design method was developed based on a position-specific score matrix (PSSM).
  • PSSM position-specific score matrix
  • a scoring scheme is used to predict the efficacy of siRNA oligos.
  • the score is a weighted sum of 39 bases (10 bases upstream of the 19mer, 19 bases on the siRNA proper, and 10 bases downstream) computed as follows:
  • a random-mutation hill climbing (RMHC) search algorithm was utilized to optimize the weights based on a training oligo set and the resulting profiles applied to a test set, with the optimizing criteria being the correlation coefficient between the knock- down (KD) levels of the oligos and the computed PSSM scores.
  • the metric to measure the effectiveness of the training and testing is the aggregate false detection rate (FDR) based on the ROC curve, and is computed as the average of the FDR scores of the top 33% oligos sorted by the scores given by the trained predictor. In computing the FDR scores, those oligos with silencing levels less than the median are considered false, and those more than the median silencing levels considered true.
  • FIG. 3 illustrates typical ROC curves, generated by an ensemble of about 200 randomly optimized predictors. It could be seen that the performance of the training is better than the test set, which is hardly surprising. Both curves are significantly better than random.
  • FIG. 5 illustrates the resulting sequence profiles from training and testing on several different oligo sets. This profile illustrates that G or C bases are strongly preferred at the beginning, i.e., the 5' end, and strongly disfavored at the end, i.e., the 3' end, of the 19mer sequence. To confirm this observation, the average knock-down levels for oligos starting and ending with G/C or AAJ are computed, and those oligos starting with G/C and ending with AAJ have the best performance, far superior to the other three categories.
  • a 19mer oligo having a sequence of GCGTTAATGTGATAATATA (SEQ ID NO: 1), and the oligos that are most similar to this sequence are identified as an siRNA that may have high silencing efficacy.
  • the design method incorporated both PSSMs shown in FIG. 3 because the combination gave better performance as compared to using either one PSSM alone.
  • the improved siRNA design method selected oligonucleotides based on 4 principles: base composition, off-target identity, position in the transcript, and sequence variety. Certain oligonucleotides containing sequence from features such as untranslated regions, repeats or homopolymeric runs were eliminated. Remaining oligonucleotides were ranked by their PSSM scores. Top-ranking oligonucleotides were selected for variety in GC content, in start position, and in the two bases upstream of the siRNA 19mer duplex.
  • oligonucleotides were then filtered for predicted off-target activity, which was calculated as a position-weighted FASTA alignment score. Remaining oligonucleotides were ranked by PSSM scores, subjected to a second round of selection for variety and finally re-ranked by their PSSM scores. The desired number of siRNAs was retained from the top of this final ranking.
  • the improved method was compared to the standard method by side-by-side testing of new siRNAs selected by each.
  • the results obtained with three siRNAs selected by each method are shown in Figure 3.
  • siRNAs designed by the improved algorithm showed better median efficacy (88%, compared to 78% for the standard method siRNA) and were more uniform in their performance.
  • Table III lists the 30 siRNAs.
  • an siRNA design with the standard method had a median silencing level of 75%.
  • 28 had silencing levels equal to or better than 75%, 26 better than or equal to 80%, and 37% better than 90%, comparing with only 10% better than 90% using the standard method.
  • Two target genes (KIF 14 and IGFlR) had been very difficult to silence by siRNAs, with previous oligos achieving only 40% to 70% and no more than 80% silencing levels in the past.
  • the 12 new oligos targeting these genes all achieved silencing of at least 80% and 6 achieved 90% levels.
  • Figure 6 shows an example of alignments of transcripts of off-target genes to the core 19mer of an siRNA oligo sequence.
  • Off-target genes were selected from the Human 25k v2.2.1 microarray by selecting for kinetic patterns of transcript abundance consistent with direct effects of siRNA oligos. Alignments were generated with FASTA and edited by hand. The black boxes and grey area demonstrate the higher level of sequence similarity in the 3 ' half of the alignment.
  • the alignment shown in Fig. 6 and similar data for other siRNAs were combined to generate a position-specific scoring matrix for use in predicting off-target effects.
  • the matrix which reflects the frequency with which each position in the oligo is found to match affected off-target transcripts, is represented in Fig. 7.
  • the position-specific scoring matrix is used to calculate scores for alignments between a candidate RNAi sequence and off-target transcript sequences. Alignments of interest are established with a low-stringency FASTA search and the score for each alignment is calculated with the Eq. 6
  • the threshold of the score was optimized to maximize the correlation between predicted and observed numbers of effects (Fig. 8).
  • the selection pipeline uses the optimized threshold to favor sequences with relatively small numbers of predicted off-target effects.
  • the curve models were generated as a sum of normal curves. Each curve represents the probability of finding a particular base in a particular region. The value at each position in the summed normal curves is the weight given to that position for the base represented by the curve. The weights for each base present at each position in each siRNA and its flanking sequences were summed to generate an siRNA's score, i.e., the score is ⁇ Wj.
  • the score calculation can also be described as the dot product of the base content in the sequence with the weights in the curve model. As such, it is one way of representing the correlation of the sequence of interest with the model.
  • Curve models can be initialized to correspond to the major peaks and valleys present in the smoothed base composition difference between good and bad siRNAs, e.g., as described in FIGS. IA-C and 5A-C.
  • the initial model can be set up for the 3-peak G/C curve model as follows: Peak l mean: 1.5 standard deviation: 2 amplitude: 0.0455 Peak 1 mean, standard deviation and amplitude are set to correspond to the peak in the mean difference in GC content between good and bad siRNAs occurring within bases -2 - 5 of the siRNA target site in Set 1 training and test sets. Peak 2 mean: 1 1 standard deviation: 0.5 amplitude: 0.0337
  • Peak 2 mean standard deviation and amplitude are set to correspond to the peak in the mean difference in GC content between good and bad siRNAs occurring within bases 10- 12 of the siRNA target site in Set 1 training and test sets.
  • Peak 3 mean 18.5 standard deviation: 4 amplitude: -0.0548
  • Peak 3 mean, standard deviation and amplitude are set to correspond to the peak in the mean difference in GC content between good and bad siRNAs occurring within bases 12-25 of the siRNA target site in Set 1 training and test sets.
  • Peak height (amplitude), center position in the sequence (mean) and width (standard deviation) of a peak in a curve model can be adjusted.
  • Curve models were optimized by adjusting the amplitude, mean and standard deviation of each peak over a preset grid of values. Curve models were optimized on several training sets and tested on several test sets, e.g., training sets and test sets as described in Table II. Each base - G/C, A or U - was optimized separately, and then combinations of optimized models were screened for best performance.
  • the optimization criteria for curve models were: (1) the fraction of good oligos in the top 10%, 15%, 20% and 33% of the scores, (2) the false detection rate at 33% and 50% of the siRNAs selected, and (3) the correlation coefficient of siRNA silencing vs. siRNA scores used as a tiebreaker.
  • a grid of possible values for amplitude, mean and standard deviation of each peak is explored.
  • the models with the top value or within the top range of values for any of the above criteria were selected and examined further.
  • G/C models were optimized with 3 or 4 peaks. A models were optimized with 3 peaks. U models were optimized with 5 peaks. Exemplary optimization ranges for the models are listed below:
  • FIG. 1 IA An exemplary set of curve models for PSSM is shown in FIG. 1 IA.
  • FIG. 1 IB shows the performance of the models on training and test sets.
  • the mean difference in G/C content between good and bad siRNAs provides a model for G/C PSSMs which can be used to classify siRNA functional and resistant motifs.
  • both strands of the siRNA can be active (see, e.g., Elbashir et al., 2001, Genes Dev. 15:188-200), it was of interest to discover how well the G/C contents of both sense and antisense strands of siRNAs fit the model of siRNA functional target motif G/C content derived from the mean difference in G/C content between good and bad siRNAs.
  • the reverse complements of good and bad siRNAs were examined. These reverse complements correspond to the hypothetical perfect match target sites for the sense strands of the siRNA duplexes.
  • FIG. 14A shows the difference between the mean G/C content of the reverse complements of bad siRNAs with the mean G/C content of the bad siRNAs themselves, within the 19mer siRNA duplex region.
  • the difference between the mean G/C content of good and bad siRNAs is shown for comparison. The curves were smoothed over a window of 5 (or portion of a window of 5, at the edges of the sequence).
  • FIG. 14B shows the difference between the mean G/C content of the reverse complements of good siRNAs with the mean G/C content of bad siRNAs, within the 19mer siRNA duplex region.
  • the difference between the mean G/C content of good and bad siRNAs is shown for comparison.
  • the curves were smoothed over a window of 5 (or portion of a window of 5, at the edges of the sequence).
  • the reverse complements of bad siRNAs were seen to be even more different from the bad siRNAs themselves than are good siRNAs.
  • the reverse complements of bad siRNAs had even stronger G/C content at the 5' end than the good siRNAs did and were similar in G/C content to good siRNAs at the 3' end.
  • the reverse complements of good siRNAs were seen to be substantially more similar to bad siRNAs than the good siRNAs were.
  • the reverse complements of good siRNAs hardly differed from bad siRNAs in G/C content at the 5' end and were only slightly less G/C rich than bad siRNAs at the 3' end.
  • siRNAs with strong sense strands are distinguishing siRNAs with strong sense strands as bad siRNAs from siRNAs with weak sense strands as good siRNAs.
  • An siRNA whose G/C PSSM score is greater than the G/C PSSM score of its reverse complement is predicted to have an antisense strand that is more active than its sense strand.
  • an siRNA whose G/C PSSM score is less than the G/C PSSM score of its reverse complement is predicted to have a sense strand that is more active than its antisense strand. It has been shown that increased efficacy corresponds to greater antisense strand activity and lesser sense strand activity.
  • G/C PSSMs of this invention would appear to distinguish good siRNAs with greater efficacy due to dominant antisense strand activity ("antisense-active" siRNAs) from siRNAs with dominant sense strand activity (“sense-active” siRNAs).
  • antisense-active siRNAs siRNAs with dominant sense strand activity
  • siRNAs and their reverse complements were scored using the smoothed G/C content difference between good and bad siRNAs within the 19mer, shown in FIG. 14 A, as the weight matrix.
  • the G/C PSSM score of each strand was the dot product of the siRNA strand G/C content with the G/C content difference matrix, following the score calculation method of curve model PSSMs.
  • siRNAs were called sense-active by the 3 '-biased method of expression profile analysis if the antisense-identical score exceeded the sense-identical score.
  • siRNAs were called sense-active by the G/C PSSM method if their reverse complement G/C PSSM score exceeded their own G/C PSSM score.
  • siRNAs were binned by measured silencing efficacy, and the frequency of sense-active calls by the expression profile and G/C PSSM methods was compared. Although these techniques are based on distinct analyses, the agreement is quite good. Both show that a higher proportion of low-silencing siRNAs vs. high-silencing siRNAs are predicted to be sense active.
  • the correlation coefficient for (siRNA G/C PSSM score - reverse complement G/C PSSM score) vs. log ! o(sense-identity score/antisense-identity score) is 0.59 for the set of 61 siRNAs binned in FIG. 15.
  • siRNA design algorithm that permits selection of siRNAs with greater and more uniform silencing ability was described. Despite this dramatic improvement, some genes remain difficult to silence with high efficacy. A general trend toward poorer silencing for poorly-expressed genes (less than - 0.5 intensity on microarray; ⁇ 5 copies per cell; Figure 16) was observed.
  • This example describes identification of parameters affecting silencing efficacy of siRNAs to poorly expressed genes. Twenty-four poorly-expressed genes were selected for detailed analysis of parameters affecting siRNA silencing efficacy. A number of criteria were evaluated for their ability to distinguish good and bad siRNAs, including base composition of the 19mer siRNA duplex sequence and the flanking target region. In addition, the contribution of the GC content of the target transcript was considered.
  • siRNA efficacy correlated well with siRNA and target gene base composition.
  • the GC content of good siRNAs differed substantially from that of bad siRNAs in a region- specific manner ( Figure 17).
  • the sequences of siRNAs used in generating Figure 17 are listed in Table IV.
  • Good siRNA duplexes tended to be GC poor at positions 2-7 of the 5' end of the sense strand, and GC poor at the 3' end (positions 18-19).
  • siRNA efficacy correlated with low GC content in the transcript sequence flanking the siRNA binding site. The requirement for low GC content as a determinant of siRNA efficacy may explain the difficulty in silencing the poorly-expressed transcripts, as these transcripts tend to be GC rich overall.
  • Base composition of the siRNA duplex also affected silencing of poorly expressed genes.
  • the GC content of good siRNAs differed substantially from that of bad siRNAs in a region-specific manner ( Figure 17).
  • Good siRNA duplexes tended to be GC rich at the first position, GC poor at positions 2-7 of the 5' end of the sense strand, and GC poor at the 3' end (positions 18-19.)
  • low GC content in positions 2-7 of the sense strand ( Figure 17, dotted line) produced the greatest improvement in silencing efficacy. This is consistent with the region of the siRNA implicated in the catalysis step of transcript silencing.
  • Low GC content in this region may provide accessibility or optimal helical geometry for enhanced cleavage. Requiring low GC content in this region of the siRNA may also select for target sites that contain low GC content flanking the binding site, which also correlated with silencing efficacy.
  • the base composition for good siRNAs to poorly-expressed genes diverges somewhat from our previously-derived base composition criteria for good siRNAs to well- expressed genes (Figure 17, solid line).
  • Good siRNAs to both types of genes show a preference for high GC at position 1, and low GC at the 3' end.
  • siRNAs for well-expressed genes show an extreme asymmetry in GC content between the two termini, while siRNAs for poorly-expressed genes prefer a more moderate asymmetry.
  • Our previous design algorithm seeks to maximize asymmetry, in accordance with the features seen in good siRNAs to well-expressed genes. Our current results indicate that base composition of more than one region of the siRNA can influence efficacy.
  • siRNA design algorithm (the third generation siRNA design algorithm, RSTA siRNA V3) was developed based on the GC composition derived for poorly- expressed genes.
  • the new algorithm includes the following adjustments to the previous algorithm: (1) selection for 1-3 G+C in sense 19mer bases 2-7,
  • siRNAs designed by the new algorithm of the present example showed better median efficacy (80%, compared to 60% for the standard method siRNA) and were more uniform in their performance.
  • siRNAs designed using the new design algorithm also appear effective at silencing more highly-expressed transcripts, based on an examination of 12 highly-expressed genes.
  • the new design criteria may capture features important to siRNA functionality in general ( Figure 19), and emphasize that different regions of siRNAs have different functions in transcript recognition, cleavage, and product release. Bases near the 5' end of the guide strand are implicated in transcript binding (both on- and off-target transcripts), and have recently been shown to be sufficient for target RNA-binding energy.
  • the design criteria are also consistent with available data on how siRNAs interact with RISC, the protein-RNA complex that mediates RNA silencing.
  • the designs described in these examples include a preference for U at position 10 of the sense strand, which has been associated with improved cleavage efficiency by RISC as it is in most endonucleases.
  • the observed preference for low GC content flanking the cleavage site may enhance accessibility of the RISC/nuclease complex for cleavage, or release of the cleaved transcript, consistent with recent studies demonstrating that base pairs formed by the central and 3' regions of the siRNA guide strand provide a helical geometry required for catalysis.
  • the new design criteria may increase the efficiency of these and additional steps in the RNAi pathway, thereby providing efficient silencing of transcripts at different levels of expression.
  • NM_0Q4 5 23 KNSLl GTTTTCTCTGTTACAATAC 433

Landscapes

  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biotechnology (AREA)
  • Chemical & Material Sciences (AREA)
  • Genetics & Genomics (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Medical Informatics (AREA)
  • Proteomics, Peptides & Aminoacids (AREA)
  • Theoretical Computer Science (AREA)
  • Analytical Chemistry (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • Wood Science & Technology (AREA)
  • Zoology (AREA)
  • Organic Chemistry (AREA)
  • Plant Pathology (AREA)
  • Microbiology (AREA)
  • Biochemistry (AREA)
  • Micro-Organisms Or Cultivation Processes Thereof (AREA)
  • Pharmaceuticals Containing Other Organic And Inorganic Compounds (AREA)

Abstract

L'invention porte sur des procédés et des appareils et des produits programme d'ordinateur pour classer des motifs de séquence fonctionnelle d'ARNsi pour une utilisation dans une séquence centrale d'ARNsh, et pour mettre au point un ARNsh, des séquences centrales d'ARNsh ou des séquences de brins sens de celles-ci ou des séquences anti-sens de celles-ci. La séquence centrale d'ARNsh comprend une séquence de brin anti-sens et une séquence de brin sens et est mise au point pour produire un ARNsi pour silencer un gène cible dans une cellule. Dans un mode de réalisation, le procédé sélectionne une première pluralité de motifs de séquence fonctionnelle d'ARNsi avec chacun une efficacité de silençage prédite contre un motif de séquence ciblé à l'intérieur d'un produit de transcription d'un gène cible qui est égal ou supérieur à un seuil choisi d'efficacité de silençage, et classifie chaque motif de séquence fonctionnelle d'ARNsi dans la première pluralité de motifs de séquence fonctionnelle d'ARNsi comme appartenant à une première classe de motifs de séquence fonctionnelle d'ARNsi lorsque le motif de séquence fonctionnelle d'ARNsi a un nucléotide 3' terminal qui est une cytosine, ou comme appartenant à une seconde classe de motifs de séquence fonctionnelle d'ARNsi lorsque ledit motif de séquence fonctionnelle d'ARNsi n'a pas un nucléotide 3' terminal qui est une cytosine. L'invention décrit en outre des procédés de mise au point par le remplacement d'un nucléotide non-cytosine 3' terminal par une cytosine ou par addition d'une cytosine 3' terminal à un motif de séquences fonctionnelle d'ARNsi.
PCT/US2008/011010 2007-09-24 2008-09-23 Procédés de mise au point d'arn en épingle à cheveux courts (arnsh) pour le silençage de gènes Ceased WO2009042115A2 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US99515607P 2007-09-24 2007-09-24
US60/995,156 2007-09-24

Publications (2)

Publication Number Publication Date
WO2009042115A2 true WO2009042115A2 (fr) 2009-04-02
WO2009042115A3 WO2009042115A3 (fr) 2009-07-30

Family

ID=40377695

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2008/011010 Ceased WO2009042115A2 (fr) 2007-09-24 2008-09-23 Procédés de mise au point d'arn en épingle à cheveux courts (arnsh) pour le silençage de gènes

Country Status (1)

Country Link
WO (1) WO2009042115A2 (fr)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022225997A1 (fr) * 2021-04-20 2022-10-27 Speratum Biopharma, Inc. Flux de travail pour conception de séquence de substrat pour la régulation multi-site médiée par arni
CN116825199A (zh) * 2023-02-21 2023-09-29 王全军 筛选siRNA序列以降低脱靶效应的方法及系统
CN120148613A (zh) * 2025-03-03 2025-06-13 寻鲸生科(北京)智能技术有限公司 基于深度学习的siRNA序列设计方法及活性预测装置

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1926551B (zh) * 2003-10-27 2010-06-16 罗斯塔生化科技有限责任公司 用于基因沉默的siRNA的设计方法
EP1979480A2 (fr) * 2005-09-12 2008-10-15 Somagenics, Inc. Inhibition de l'expression genique virale a l'aide d'un petit arn interferent

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022225997A1 (fr) * 2021-04-20 2022-10-27 Speratum Biopharma, Inc. Flux de travail pour conception de séquence de substrat pour la régulation multi-site médiée par arni
CN116825199A (zh) * 2023-02-21 2023-09-29 王全军 筛选siRNA序列以降低脱靶效应的方法及系统
CN120148613A (zh) * 2025-03-03 2025-06-13 寻鲸生科(北京)智能技术有限公司 基于深度学习的siRNA序列设计方法及活性预测装置

Also Published As

Publication number Publication date
WO2009042115A3 (fr) 2009-07-30

Similar Documents

Publication Publication Date Title
AU2004286261B2 (en) Method of designing siRNAs for gene silencing
Shao et al. Effect of target secondary structure on RNAi efficiency
Knott et al. A computational algorithm to predict shRNA potency
Huppi et al. Defining and assaying RNAi in mammalian cells
Ui-Tei et al. Functional dissection of siRNA sequence by systematic DNA substitution: modified siRNA with a DNA seed arm is a powerful tool for mammalian gene silencing with significantly reduced off-target effect
Kamola et al. The siRNA non-seed region and its target sequences are auxiliary determinants of off-target effects
Pei et al. On the art of identifying effective and specific siRNAs
Khvorova et al. Functional siRNAs and miRNAs exhibit strand bias
Sætrom et al. A comparison of siRNA efficacy predictors
Matveeva et al. Comparison of approaches for rational siRNA design leading to a new efficient and transparent method
Laganà et al. Computational design of artificial RNA molecules for gene regulation
Rice et al. Functional atlas of primary miRNA maturation by the microprocessor
Horn et al. Design and evaluation of genome-wide libraries for RNA interference screens
Okamura et al. The regulatory activity of microRNA* species has substantial influence on microRNA and 3′ UTR evolution
Ichihara et al. Thermodynamic instability of siRNA duplex is a prerequisite for dependable prediction of siRNA activities
Cullen Induction of stable RNA interference in mammalian cells
Naito et al. Designing functional siRNA with reduced off-target effects
Yan et al. Improving the prediction of human microRNA target genes by using ensemble algorithm
Hansen et al. Enhancing miRNA annotation confidence in miRBase by continuous cross dataset analysis
Bofill-De Ros et al. Flexible pri-miRNA structures enable tunable production of 5’isomiRs
JP2006525811A (ja) Ｒｎａ干渉の方法と組成物
Arroyo et al. Systematic design and functional analysis of artificial microRNAs
Liu et al. The effect of regions flanking target site on siRNA potency
WO2009042115A2 (fr) Procédés de mise au point d'arn en épingle à cheveux courts (arnsh) pour le silençage de gènes
US20140088937A1 (en) Methods of Predicting The Probability of Modulation of Transcript Levels By RNAI Compounds

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 08832903

Country of ref document: EP

Kind code of ref document: A2

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 08832903

Country of ref document: EP

Kind code of ref document: A2