JPH0694696A - Chromatogram analysis method and chromatograph device - Google Patents

Abstract

(57) [Summary] [Purpose] Accurately and quantitatively calculate each peak from overlapping chromatogram peaks by high-speed data processing. [Construction] A neural network type peak area distribution function 61 is provided in the data analysis unit 45, and the respective areas of the chromatogram peaks in which the unknown sample is overlapped are output using the coupling constant (process processing constant) of this neural network. In addition to the peak overlap, the baseline can be determined using the coupling constant of the neural network from the peak overlap with the baseline. [Effect] In the glycohemoglobin analyzer, l-
Even if Alc and s-Alc are not sufficiently separated to form a shoulder peak, the quantification accuracy is 3 to 5% for s-Alc,
With l-Alc, the error can be suppressed to about 10%.

Description

Detailed Description of the Invention

[0001]

FIELD OF THE INVENTION The present invention relates to the analysis of chromatograms obtained from chromatographs.

[0002]

2. Description of the Related Art Conventionally, when obtaining the area, height, etc. of each peak from the overlapping of insufficiently separated peaks in a chromatogram, the peaks between the peaks (local minimum points) may be divided vertically with respect to the time axis. Alternatively, it is divided so that a sub-peak from the tailing peak and the leading peak is cut out by a straight line. This method is simple, but lacks the accuracy of peak quantification, and is often biased by ± 50% from the true value.

As a method of improving this, Gaussi
an (G function) and Exponentially mod
A technique for fitting overlapping peaks by using a nonlinear least squares method using an analytical function such as if-defined Gaussian (EMG function), and further, a nonlinear minimum 2
It has been proposed to use data of a three-dimensional chromatogram having three axes of time, wavelength, and absorbance (intensity) of the chromatogram as the data when performing the peak fitting by the multiplication.

[0004]

However, when the least squares method as described above is used, the optimum parameters used for the nonlinear least squares method cannot be obtained as expected, and the solution may not converge. In the latter case, when using a three-dimensional chromatogram, it is troublesome to obtain three-dimensional information. Furthermore, if there is no difference in wavelength spectrum between peaks, fitting is considerably difficult.

The present invention has been made in view of the above points, and an object of the first invention is to obtain information on two-dimensional chromatograms on a time axis and an intensity axis, and even if overlapping peak patterns are various, Chromatogram analysis method and chromatograph device capable of accurately grasping features (area ratio, peak height, retention time, etc.), suppressing quantitative error of each peak to about several percent, and capable of high-speed data processing To provide.

A second object of the present invention is to accurately determine the baseline of each chromatogram even if the baseline and the peak overlap exist in various patterns, and the accuracy of the determination is high. It is an object of the present invention to provide a chromatogram analysis method and a chromatograph device that enable peak quantification.

[0007]

In order to achieve the first object, the present invention proposes the following chromatogram analysis method.

That is, by using a peak shape function, overlapping peaks of various patterns having different areas, heights, retention times, etc. of overlapping peaks are simulated in advance,
From these simulated data, the area, height, or retention time of each overlapped peak is established as a solution processing constant for the black box, and this processing processing constant is used to overlap the unknown chromatograms input to the black box. It is applied to peak data analysis (this is the 1-1st problem solving means).

Alternatively, instead of obtaining a peak shape of a chromatogram in a known sample such as a standard sample, an unknown sample (here, the unknown sample is a general sample composed of the same components as the standard sample and other components). If there is an overlap between the peaks of the unknown sample and the peak of the unknown sample, the peak area S1 (or peak waveform P1) of the standard sample, the height h1, and the total of the overlapping peaks of the unknown sample are obtained. From the area S2 (or overlapping waveform P2) and the peak height h2 of the same component as the standard sample, the peak area S3 (or peak waveform P3) of the other component of the unknown sample is obtained.
S3 = S2-S1 · h2 / h1 or P3 = P2-
A method of calculating from the relational expression of P1 · h2 / h1 is proposed (this will be referred to as the 1-2nd problem solving means).

Further, in order to achieve the second object, various patterns in which a baseline and a peak are combined are simulated in advance, and the baseline mode can be derived as a solution from these simulated data. And apply this process processing constant to the data analysis of the baseline determination of the unknown chromatogram input to the black box (this is the second means for solving the problem).

Incidentally, as the process processing constants in the 1-1 th problem solving means and the second problem solving means, a coupling constant by a neural network is typical,
This is described in detail in the examples.

[0012]

Operation of the 1-1st problem-solving means: Chromatograms of overlapping peaks of various patterns are simulated in advance by using peak shape functions. In this case, it is possible to select the optimum overlapping pattern of the chromatograms to be simulated by trial and error of the parameters to be substituted into the peak shape function, by the data creator. As a result, various peak overlapping chromatogram patterns (simulated data) whose peak area, peak height, retention time, etc. are known can be obtained. Although this peak shape function uses a G function, an EMG function, or the like, it is not used for peak fitting of an actual chromatogram as in the conventional case, but a simulation data for establishing a process processing constant described below is used. Used as a creation.

A process processing constant is a solution of the data (here, the area, height, and retention time of overlapping peaks) when data (here, chromatogram of overlapping peaks) is input to a black box as a data processing means. To guide to the output side.

As for this process processing constant, if the simulation data that the solution knows beforehand is used, the learning function can be used to obtain the process processing constant value at which a known solution is output with respect to the input simulation data. The more various kinds of data exist, the more various process processing constants adapted to them can be established. The process processing information thus obtained is stored. When an actual unknown chromatogram is entered in the black box, the process processing constant of simulated data close to this chromatogram is applied to analyze overlapping peak data (for example, area ratio between peaks, peak height, The retention time, etc.) is calculated, and it becomes possible to quantitatively calculate each peak.

In this case, since the process processing constant of the simulated data which is the closest to the actual data analysis of the chromatogram is used, a large error does not occur in the data analysis,
Moreover, it can be applied to data analysis of various patterns.

1-2 Operation of means for solving problems: When the peak area of the standard sample is S1 and the height is h1, while the total area of overlapping peaks of the unknown sample is S2 and the peak height of the standard sample is h2. , S1 · h2 / h1 corresponds to the peak area of the same component as the standard sample in the unknown sample. Therefore, from the total area S2 of the peaks of the unknown sample to S1 · h2 / h
If 1 is subtracted, the peak area S of the remaining components of the unknown sample
3 can be obtained easily and with a small error. According to this method, the peak area S1 is replaced with the peak waveform P1, the total area S2 of overlapping peaks is replaced with the overlapping peak waveform P2, and the peak area S3 is replaced with the peak waveform P3, whereby it becomes possible to quantitatively calculate each peak of the unknown sample. .

Operation of the second means for solving the problems ... In this method,
Instead of overlapping the peaks of the first means for solving problems, the data analysis target was the baseline of the peaks. In the case of the present invention, since the baseline as a solution is known in the simulated chromatogram, the process processing constant of the black box that leads this solution is also obtained by inputting the simulated data into the black box and learning. be able to.

By preparing various kinds of this baseline and peak simulation data, it is possible to prepare various kinds of process processing constants corresponding thereto.

When an actual unknown chromatogram is input to the black box, process processing constants of simulated data close to this chromatogram are applied to accurately determine the baseline of the chromatogram, and to quantify each peak. The calculation can be done accurately.

[0020]

Embodiments of the present invention will be described with reference to FIGS.

First, the construction of the chromatographic apparatus to which the present invention is applied will be described with reference to FIG. FIG. 4 is a system configuration diagram of a liquid chromatograph for analyzing glycohemoglobin by ion exchange chromatography.

The eluent liquid feed pump 40 performs the stepwise elution with the eluent A48, the eluent B49, and the eluent C.
50 for 1.9, 1.0, 0.4 minutes each.
The solution is switched and transferred in a cycle of 3 minutes. The sampler 43 having a movable needle is a standard sample 5 of hemoglobin (Hb).
5 or 5 μl of 2 or unknown sample 51 is diluted by hemolysis,
The hemolytically diluted sample is sent to the separation column 41 together with the eluent A48, the contained components are separated and developed, and are detected by the visible light absorption detector 42. The chromatogram that is the detection data is stored in the data analysis unit 45.

The data analysis unit 45 has a data analysis function for performing peak identification and quantitative calculation for a chromatogram according to the flow chart of FIG. 1 described later, and as shown in FIG. 5, a neural network type peak area distribution function 61. , A peak detection function 62, a baseline determination function 63, a peak identification function 64, a differential calculation function 65, a chromatogram normalization function 66, and a quantitative calculation function 67. These functions will be described in the flowchart of FIG.

FIG. 2 shows an example of the chromatogram obtained by the separation column 41. When the peaks of 1-Alc (unstable hemoglobin) and s-Alc (stable hemoglobin) overlap in this way, It is difficult to measure each peak area accurately. In this embodiment, each peak area is measured by using a neural network in order to ensure the accuracy of this measurement.

Now, the chromatogram analysis executed by the data analysis unit 45 will be described with reference to the flowchart of FIG.

This flow starts when a digitized chromatogram is obtained (step 1), and in step 2, the peak of the chromatogram is detected. In the peak detection, the peak detection function 62 sets the first point at which the rising increase amount of each peak exceeds five times the noise value three times as a peak start point and vice versa as the peak end point. In step 3, the baseline determination function 63 draws a straight line so that the start point of the first peak of the chromatogram, the start point of the last peak up to 2.15 min, and the end point of the last peak are not exceeded in the chromatogram. tie. For example, as shown in FIG. 2, a straight line connects the starting point of Ala, the ending point of s-Alc, and the ending point of Ao. In step 4, peak identification is performed.
In peak identification, if there is a retention time (time of peak maximum point) in a time window (peak identification allowable time width) prepared in advance by the peak identification function 64, it is fixed to the component of that time window. If there is a maximum point (maximum point of each peak) on the chromatogram, peak identification is completed here.

As shown in FIG. 6, the s-Alc peak has l
When -Alc overlaps as a shoulder peak, identification cannot be performed by the maximum point identification method. In this case, therefore, the peak identification function 64 performs peak identification to the differential calculation function 65 when l-Alc is not identified. To order. The peak identification function 64 has an inflection point (a point where the second derivative is 0 and the first derivative is downwardly convex) that is convex downward from the peak maximum point to the left on the rising side (left side of the peak) of s-Alc. Search for it. If there is, it is identified as l-Alc when any inflection point is within the time window. Without it, we won't identify
After step 5, the process proceeds to step 17. Here s-
The quantitative calculation function 67 quantifies Alc as a single peak from the peak area, and then ends in step 18.

If the identification of l-Alc and s-Alc is confirmed in step 5, the peaks Ala, Alb, F, l-Alc, s-Alc, A identified in step 6 and thereafter are confirmed.
Quantitative calculation is performed by obtaining the peak area of o. In this quantitative calculation, the correction factor is multiplied by each peak area, and the area percentage of each peak is quantified by the quantitative calculation function 67.

The following description focuses on the determination of 1-Alc and s-Alc.

In step 6, when peak separation is performed, a. Vertical division or b. It is selected whether or not to use the neural network, but this selection is made by the operator in advance.

If vertical division is selected, it is checked in step 7 whether or not there is a valley (minimum point) behind 1-Alc. If there is, the peak area from the valley is divided perpendicularly to the time axis in step 8, and then quantitative calculation is performed in step 10 and the process ends in step 18.

Next, the case where the quantitative mode by the neural network is selected in step 6 will be described. Step 11 of the neural network flow
To step 15 from.

In step 11, a valley (minimum point) is formed in the section from the peak start point of l-Alc to the end point of s-Alc peak.
Check if is one. If there are two or more, step 1
In step 6, the CRT 46 and the printer 47 output a message such as "Unable to quantify neuron", indicating that the determination is impossible, that is, the determination by the neural network is impossible, and the process proceeds to the vertical division processing system after step 6. 0
In the case of a single piece, the first derivative of l-Alc is an inflection point that is convex downward, and this inflection point is s-Al from the peak start point of l-Alc.
Check whether there is one in the section up to the peak maximum point of c. Then, if there is one inflection point that is convex in the valley or downward, the process proceeds to step 12.

In step 12, the chromatogram normalization function 12 uses both the vertical axis and the horizontal axis to quantify the overlapping peaks of 1-Alc and s-Alc by a neural network.

To

Normalize within the interval [0, 1]. That is, FIG.
As shown in (a), 1-Al of the actual chromatogram
Start point A of c, end point B of s-Alc, l-Alc or s-A
Project the maximum point C of lc to the normalized chromatogram. A
Is set to (0.0,0.0), B is set to (1.0,0.0), and C is set to 1.0 on the vertical axis. When point A is lifted up from the baseline as shown in FIG. 3, the hatched portion is subtracted before projection. CRT this normalized chromatogram
46 and the printer 47.

Here, in step 13, it is determined whether the peak shape is suitable for the neural network, and the total peak area of l-Alc and s-Alc in the normalized chromatogram is 0.2 or more and 0.8 or less. Confirm that it is. If it is less than 0.2, the peak is sufficiently sharp and the peak is completely divided. Therefore, it is unnecessary to determine the neural amount in step 16, and if it exceeds 0.8, it is abnormally broad. It is output that the network cannot be quantified, and the process proceeds to step 7 to enter the vertical division quantitative calculation mode. Total peak area is 0.2
If it is in the range of 0.8, the process proceeds to step 14.

In step 14, the neural network type peak area distribution function 61 controls the normalized chromatogram l.
Calculate what percentage of -Alc and s-Alc is occupied to what percentage. In step 15, the area ratio of the obtained chromatogram is used for distribution and quantitative calculation is performed. That is, as shown in FIG. 3, when the point A is lifted from the baseline, the total peak area above the baseline is distributed at the area ratio obtained by the neural network for convenience. The area ratio obtained here is output to the CRT 46 and the printer 47. After that, the area percentage is calculated using the peak areas of Ala to Ao as in the case of the normal processing, and this flow is ended in step 9.

The neural network type peak area distribution function 61 will be described in detail with reference to FIG.

As shown in FIG. 5, this neural network is of the Rumelhart type and has a peak area distribution function 61 of a neural network type which is a black box.
Has a hierarchical structure consisting of an input layer, an intermediate layer, and an output layer. By adopting a supervised learning method by backprogation, the learning data (supervised data) whose answers are known in advance is input to the input layer. Then, the intermediate layer sequentially calculates the coupling constant group (process processing constant) that guides the answer to the output layer by the learning function. The output is [0,
1] is a sigmoid function S (x)
= 1 / (1 + exp (-x)) is used.

In this neural network, the coupling constant group that connects the input layer and the intermediate layer and the intermediate layer and the output layer are sequentially corrected using the learning simulation data as shown in FIG. 3 (b). .

A specific learning example for calculating the coupling constant group will be described with reference to FIG.

The learning simulation data is a chromatogram having overlapping peaks, and can be created by simulating using a peak shape function, for example, and can be simply described by the G function of the equation 1 and the EMG function of the equation 2 Can be created using.

[0043]

[Equation 1]

[0044]

[Equation 2]

That is, 1 × 1 of the normalized chromatogram
T _R (retention time), σ for each peak of l-Alc and s-Alc in the square of
(Peak width), τ (tailing peak), peak height h
3 and the like (parameters used in the above equation 1 or equation 2) are randomly varied within an appropriate interval,
As shown in (b), for example, the peak area ratio is 10: 9.
Dozens of learning simulation data of various overlapping patterns such as 0, 20:80, 30:70, 40:60 are created. However, since 1-Alc and s-Alc are close to each other in the actual chromatogram, σ and τ can be approximated by common values. In addition, the peak start point and the peak end point are determined as a height of 0.01 on the normalized intensity axis for convenience.

These dozens of types of learning simulation data are interpolated and input to the input layer as teacher data. This input is performed by dividing the section from A to B of the simulation data into 16 with respect to the time axis coordinates and giving 16 intensities at the time of this division. In the output layer, the area ratio (solution) of s-Alc corresponding to the teacher data is shown in step 1 of FIG.
A value of 0.2 to 0.8 is presented corresponding to 3. By providing such supervised data, the intermediate layer is successively corrected by learning for each input data while the coupling constant group for deriving the answer is learned.

The learning simulation data, which is the basis for calculating the above-mentioned coupling constant, can be simulated by other than an analytical function. For example, first, the peak shape of s-Alc of the standard sample is extracted as a function f (t) from the actual chromatogram. This is 1-for standard samples
This is because Alc is hardly contained. The retention time t _R and the peak width w are adopted as characteristic parameters of this function. For example, w is defined as an equation 3 using the peak area A and the peak height h.

[0048]

[Equation 3]

This function form is normalized to an area of 1, and t _R and w
Is expressed as f (t) having a constant, it corresponds to Gaussian G (t) having t _R and σ as constants. After that, by using the function f (t) of the actual peak shape in the above-described method, the time scale and the intensity scale can be expanded or contracted to be simulated.

The parameters σ, τ, and w relating to the peak width described above are treated as a function having the retention time of the peak as a variable, for example, σ (t _R ) = at _R + b when simulating more precisely. There is a need.

The coupling constant of the neural network is obtained through such learning and the data analysis unit 4 of the chromatograph is used.
It is stored in advance in 5. The learning simulation and the calculation of the coupling constant may be performed on the factory side of the chromatograph device production before collection, or may be performed on the user side as necessary. In addition, parallel distributed processing (PDP; Parallel Distributed) is applied to the chromatograph.
If a computer dedicated to Processing is installed, when the first standard sample of each analysis is measured, the training data for learning is simulated using the function f (t) of the actual peak shape, and If input to the neural network type peak area distribution function 61, the coupling constant can be updated each time.

The data analysis unit 45 calculates such a coupling constant group.
If an unknown peak overlap chromatogram is input to the input layer, the peak area ratio in step 14 of FIG. 1 can be obtained from the coupling constant group that is close to it.

Here, the normalization method of the normalization function 66 will be supplementarily described. In the embodiment described above, the start point and the end point of the overlapping peaks are linearly converted into (0,0) and (1,0) on the horizontal axis, and the maximum point of the overlapping peaks is normalized to the height of the vertical axis 1. . This is done because it is processed by a neural network, but other normalization methods can be used. That is, the horizontal axis is linearly converted in the same manner as above, and the vertical axis direction is normalized so that the total area of overlapping peaks becomes 1. In this case, the height of the normalized peak serves as an index and can be used for determining whether or not the neural network quantification is unnecessary.

For other normalization methods, the actual time axis scale is directly used for the horizontal axis and the peak area or height is set to 1
It can also be normalized to This has the advantage that interpolation processing is not required, but it is necessary to diversify the teaching data. In addition to this, there is a method in which both the total area and the maximum point of the peak are normalized to 1 to make the horizontal axis flexible. This has the characteristic that it is not necessary to determine ambiguous points that are difficult to define mathematically, such as the start and end points of a peak.

In the above-mentioned neural network type peak area distribution function 61, the pattern of 16 points was input to the input layer, but there is a method of inputting the feature amount of the pattern.
That is, from the normalized chromatogram of FIG. 3, the maximum point, the maximum point, the minimum point (valley) and the coordinates (x,
You can also enter a numerical value for y). This method is characterized in that the number of input layer and intermediate layer units can be reduced to a minimum.

As described above, it is possible to quantify using a neural network of a set of coupling constants. However, the patterns of the chromatograms are classified into classes, and a neural network having a coupling constant dedicated to each class is used. It is possible to quantify more efficiently and accurately by corresponding to. For example, classify with valleys and without valleys, or by area of normalized chromatograms,
Class 1 is 0.2 or more and less than 0.4, and 0.4 or more and 0.
Class 2 is less than 6 and class 3 is less than 0.6 and less than 0.8. In general chromatograms, this classification method is effective for quantification of shoulder peaks, quantification when small peaks overlap tailing peaks, treatment with leading peaks, and the like.

In FIG. 4, 44 is a control unit for controlling the entire chromatograph, 53 is an analysis condition determination unit, and the analysis condition determination unit 53 is the switching time of the eluent when the chromatogram cannot be quantified by the above method. The control unit 44 is instructed to correct or the like, and the feedback for improving the separation is enabled.

In the above embodiment, as output items, not only the area ratio described above, but also characteristic quantities describing a chromatogram such as peak height and retention time can be set.

According to this example, the following effects were obtained.
In the separation of 1-Alc and s-Alc in the glycohemoglobin analysis, it took about 7 minutes to obtain accurate quantification by the conventional vertical peak splitting method, but in the present embodiment, about half was required, which was the same in 3.3 minutes. It became possible to quantify.

In the 3.3 minute analysis, the quantification accuracy in the conventional vertical division method was 20 to 30 in s-Alc.
%, 1-Alc contained an error of about 50%, but in the quantitative method using this neural network, s-Al
The error was suppressed to 3 to 5% for c and about 10% for 1-Alc.

Next, another embodiment of the present invention will be described with reference to FIG.

In this embodiment, the area ratio of the overlapping peaks of the chromatogram is obtained by using the actual s-Alc table pattern of the standard sample without using the neural network method.

That is, first, as shown in FIG. 7 (a), an actual standard sample s-Alc is analyzed by a chromatograph, and its chromatogram is stored as a table pattern. H1
And area S1 are calculated. Next, as shown in FIG. 7B, a chromatogram of the unknown sample is obtained, and the total area S2 of the peak h2 of s-Alc and the overlapping peak (overlapping peak of s-Alc and l-Alc) is calculated. . When the peak of s-Alc is small and does not reach the apex, the height of the inflection point is set to h2. Then, in order to make the peak area S1 of s-Alc of the standard sample correspond to the area S1 'of s-Alc of the actual unknown sample, the ratio h2 / h2 of the peak heights h2 and h1.
The area S1 is multiplied by h1 (S1 ′ = S1 × h2 / h1),
As a result, the peak area S1 'of s-Alc of the unknown sample can be obtained, and the peak area of l-Alc of the unknown sample can be obtained from S-S1'.

The above equations can be summarized as the following equations.

[0065]

## EQU00004 ## S3 = S2-S1.multidot.h2 / h1 Specifically, as shown in FIG. 7, the numerical value is expressed in terms of electric quantity, where h1 is 340 .mu.V and S1 is 1,700,000 .mu.
Vs, h2 is 470 μV, S2 is 2.8 million
If μV · s, then by substituting in Equation 4, S3 becomes 4
It becomes 50,000 μV · s.

This method obtains the area ratio of overlapping peaks by numerical calculation, but can be developed to a method of subtracting the waveform of the chromatogram. That is, the waveform of the standard sample obtained by multiplying the waveform of the general sample by a certain constant is subtracted, and l
The area value of -Alc and s-Alc can be calculated.

If this is expressed by a formula, it can be expressed as follows.

[0068]

[Equation 5] P3 = P2-P1 · h2 / h1 where P1 is the peak waveform of the standard sample, P2 is the overlapping peak waveform of the general sample, and P3 is the peak waveform of the component other than the same component as the standard sample in the general sample. Is.

In this case, h2 / h1 is almost appropriate as the constant, but some adjustment may be made so as to satisfy the following condition.

(1) Even a part of the subtracted waveform does not fall below the baseline. (2) The subtracted waveform has one vertex and one inflection point to the left and right of the vertex (first In the case where there is no influence of the peak of 3) Finally, an application example other than the glycohemoglobin analysis which has been the example so far will be described.

In catecholamine analysis, when a biological fluid such as plasma or urine is used as a sample, an interfering component often overlaps with a target component. In this case, quantification based on the peak height is less susceptible to interference. Even so, if the overlapping manner is considerably large like DA (dopamine) in FIG. 8C, the quantitative component becomes large. Also in this case, the necessary component and the disturbing component are simulated using the peak shape function as in the first embodiment, and the neural network is used to input the normalized chromatogram pattern to the input layer and the normalized chromatogram to the output layer. A method of corresponding the true value of the peak height is effective. Even in amino acid analysis, small peaks may overlap the shoulder peak and tailing peak. This can also be accurately quantified by obtaining the peak area ratio using the neural network described above.

Amino acid analysis also has a characteristic baseline variation. Cy as shown in FIGS. 8B and 8C
There is a plateau near NH ₃ (ammonia) due to the eluent switching shock immediately before s (cysteine). Since the size and shape of this type of baseline significantly changes in each chromatogram, the way of drawing the baseline is affected, and an error occurs in the quantification of peaks. For this, it is effective to use a neural network to learn some patterns and draw a baseline according to the teacher data.

That is, various patterns in which the baseline and the peak are combined are simulated in advance, a neural network is used from these simulated data to establish a group of coupling constants that can be derived as a solution of the baseline mode, and these coupling constants are Applied to data analysis of baseline determination of unknown chromatograms entered using.

[0074]

As described above, according to the present invention,
In the first and second problem solving means, the time axis and the strength axis are divided into two.
Even if the pattern of overlapping peaks varies from the data of the three-dimensional chromatogram, the features of each peak can be accurately detected, the quantification error of each peak can be minimized, and accurate quantification can be performed by high-speed data processing. .

Further, according to the second means for solving problems, even for chromatograms including various baselines, the baselines can be accurately determined to enable accurate peak quantification.

[Brief description of drawings]

FIG. 1 is a flowchart of chromatogram analysis according to the first embodiment of the present invention.

FIG. 2 is a diagram showing an example of a chromatogram of a quantification target.

FIG. 3 is an explanatory diagram of processing of a neural network.

FIG. 4 is a configuration diagram of a chromatograph device used in the first embodiment.

5 is a system configuration diagram of the data analysis unit used in FIG.

FIG. 6 is a diagram showing an example of a shoulder peak.

FIG. 7 is an explanatory diagram showing a chromatogram analysis process according to the second embodiment of the present invention.

FIG. 8 is an explanatory diagram showing a chromatogram analysis process according to the third embodiment of the present invention.

[Explanation of symbols]

40 ... Liquid feed pump, 41 ... Separation column, 42 ... Detector,
43 ... Sampler, 44 ... Control unit, 45 ... Data analysis unit,
61 ... Neural network type distribution function.

Front page continuation (72) Inventor Hiroshi Satake 882 Ichige Ichi, Katsuta, Ibaraki Hitachi, Ltd. Measuring Instruments Division (72) Inventor Yoshiaki Seki 832 Nagakubo, Horiguchi, Katsuta City, Ibaraki Prefecture

Claims (8)

[Claims] 1. A peak shape function is used to preliminarily simulate overlapping peaks of various patterns in which areas, heights, retention times, etc. of the overlapping peaks are changed, and the overlapping peaks are obtained from these simulated data. The process processing constant of the black box that can be derived as the solution of the area, height, or retention time of is established, and this processing process constant is applied to the data analysis of the overlapping peak of the unknown chromatogram input to the black box. Chromatogram analysis method. 2. The shape function of the peak according to claim 1, wherein the peak shape function is Gaussian, Exponentially.
A chromatogram analysis method, which is an analytical function such as modified Gaussian (EMG) or a shape function of a peak obtained from an actual chromatogram. 3. A variety of patterns in which a baseline and a peak are combined are simulated in advance, and a black box process treatment constant that can be derived as a solution of the baseline mode is established from these simulated data, and this process treatment constant is A chromatogram analysis method, which is applied to data analysis of baseline determination of an unknown chromatogram input to a black box. 4. The chromatogram analysis method according to claim 1, wherein the process processing constants are a group of coupling constants calculated using a neural network. 5. The chromatogram peak shape of a known sample such as a standard sample is obtained, and the chromatogram of an unknown sample (where the unknown sample is a general sample composed of the same components as the standard sample and other components). When there is an overlap between the peaks of the unknown sample, the peak area S1, height h1 of the standard sample, the total area S2 of overlapping peaks of the unknown sample, and the peak height of the same component as the standard sample The peak area S3 of the other component of the unknown sample is calculated from the length h2 by a relational expression of S3 = S2-S1 · h2 / h1. 6. A chromatogram of a chromatogram in a known sample such as a standard sample is obtained, and a chromatogram in an unknown sample (where the unknown sample is a general sample composed of the same components as the standard sample and other components). When the peak shape is obtained and the peaks of the unknown sample overlap, the peak waveform P1 of the standard sample, the height h1, and the overlapping peak waveform P2 of the unknown sample, the peak height of the same component as the standard sample From h2, the peak waveform P of the other components of the unknown sample
3 is calculated from the relational expression of P3 = P2-P1 · h2 / h1. 7. A chromatographic apparatus comprising a function for performing quantitative calculation using the chromatogram analysis method according to claim 1. Description: 8. A chromatograph device according to claim 7, wherein the parallel distributed processing system is mounted.