Data processing method and data processing system

Abstract

To highly accurately determine the quantity of a component with a relatively low concentration even when the peaks of a plurality of components accounting for the extremely large relative concentration ratio in a sample overlap each other on a chromatogram.SOLUTION: A data processing method is to, by using actual data of a three-dimensional chromatogram consisting of a chromatogram and a spectrum acquired by chromatography analysis on a sample, separate the peaks of a plurality of components overlapping on the chromatogram, and the method includes: a peak-to-be-adjusted acquisition step of applying a peak model function prepared in advance to approximate the waveform of the chromatogram to the chromatogram to obtain a plurality of peaks to be adjusted; and a peak-to-be-adjusted adjustment step of taking the plurality of peaks to be adjusted obtained in the peak-to-be-adjusted acquisition step as initial values before adjustment, and repeating the adjustment of the peaks to be adjusted until pseudo data of the three-dimensional chromatogram for the sample obtained by combining the peaks to be adjusted after adjustment becomes similar to the actual data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a data processing method and a data processing system using three-dimensional chromatogram data. [Background technology]

[0002] In liquid chromatography (LC) using a multichannel detector such as a photodiode array (PDA) detector, three-dimensional chromatogram data having three dimensions, time, wavelength, and signal intensity (absorbance), can be obtained by continuously acquiring the absorption spectrum of a sample eluting from an analytical column.

[0003] When quantifying a target component in a sample using a liquid chromatograph, it is common to create a chromatogram using the wavelength at which the target component has the greatest absorbance, and then determine the area value of the peak of the target component on the chromatogram to perform the quantification. However, a sample may contain impurities other than the target component, and the peak of the impurity may overlap with the peak of the target component on the chromatogram. In such a case, if the multiple peaks remain overlapped, the peak area values ​​of the target component and impurities cannot be obtained, and quantitative results cannot be obtained, so it is necessary to separate the multiple components whose peaks overlap on the chromatogram from each other.

[0004] Known algorithms for separating overlapping peaks (called peak separation algorithms) include an algorithm that estimates the chromatograms and spectra of multiple components whose peaks overlap by applying a peak model function such as an EMG (Exponential Modified Gaussian) function to the waveform of an actual chromatogram (see Patent Document 1). [Prior art documents] [Patent documents]

[0005] [Patent Document 1] International Publication No. 2016 / 035167 Summary of the Invention [Problem to be solved by the invention]

[0006] Among peak separation algorithms using peak model function fitting (hereinafter referred to as model-based algorithms), there are some that use an improved EMG function that can also express the tailing and leading of the actual peak waveform, and the waveform shape of the actual chromatogram can be reproduced with high accuracy by the improved EMG function, and as a result, each component can be quantified with high accuracy. However, when peaks of multiple components with extremely large relative concentration ratios in a sample, such as 100:0.05, overlap each other on a chromatogram, it has been found that when the model-based algorithm is applied to separate the multiple peaks, the peak area values ​​of low-concentration components are significantly different from the actual peak area values.

[0007] The present invention has been made in consideration of the above problems, and aims to enable highly accurate quantification of components with relatively low concentrations even when peaks of multiple components with extremely large relative concentration ratios in a sample overlap each other on a chromatogram. [Means for solving the problem]

[0008] In the model-based algorithm, the shape and size of multiple peaks overlapping each other on the chromatogram are estimated by fitting a prepared peak model function to the actual chromatogram waveform while adjusting its parameters (e.g., height, spread). Therefore, the estimated peak shape of each component is restricted by the peak model function. Therefore, there is a slight deviation between the area value of each peak replaced by the peak model function and the area value of each actual peak. This is thought to be the reason why the quantitative accuracy of a component with a relatively low concentration deteriorates when separating and quantifying peaks of multiple components with an extremely high relative concentration ratio, such as 100:0.05.

[0009] Incidentally, there are also peak separation algorithms that do not use a peak model function (hereinafter referred to as model-free algorithms). Representative examples of model-free algorithms include those that use matrix decomposition such as NMF (non-negative matrix factorization). Model-free algorithms that use matrix decomposition are algorithms that separate an original three-dimensional chromatogram into a number of peaks specified by a mathematical method, and fine-tune each peak after separation so that a pseudo three-dimensional chromatogram obtained by synthesizing each peak after separation is close to the original three-dimensional chromatogram. Such model-free algorithms have a high degree of freedom in peak separation because the peak shape after separation is not restricted by the peak model function, and it is possible to make the pseudo three-dimensional chromatogram obtained by synthesizing each peak after separation as close as possible to the original three-dimensional chromatogram. On the other hand, since there is no information about the peak shape after separation instead of being restricted by the peak model function, the shape of each peak after separation may be completely different from the actual peak shape. Therefore, the peak separation results by the model-free algorithm often have a worse reproducibility than the peak separation results by the model-using algorithm.

[0010] The inventors of the present application focused on the high degree of freedom of peak separation in the model-free algorithm and came up with the idea of ​​complementing the model-based algorithm with a model-free algorithm. The inventors of the present application then discovered that by using the peak separation results obtained using the model-based algorithm as initial data and fine-tuning the model-free algorithm, good separation accuracy can be obtained even when the peaks of multiple components with extremely large relative concentration ratios, such as 100:0.05, overlap. The present invention was made based on this discovery.

[0011] The data processing method according to the present invention is a data processing method for separating peaks of multiple components overlapping on a chromatogram by using actual data of a three-dimensional chromatogram consisting of a chromatogram and a spectrum obtained by chromatographic analysis of a sample, and includes: an adjustment target peak acquisition step for obtaining multiple adjustment target peaks by applying a peak model function prepared in advance to the chromatogram in order to approximate the waveform of the chromatogram to the chromatogram; an estimated data creation step for synthesizing the multiple adjustment target peaks obtained in the adjustment target peak acquisition step to generate estimated data of a chromatogram and a spectrum for each of the multiple components; and an adjustment target peak adjustment step for setting the estimated data created in the estimated data creation step as an initial value before adjustment, and repeating adjustment of the adjustment target peak until pseudo data of a three-dimensional chromatogram for the sample obtained by synthesizing the adjusted adjustment target peaks becomes similar to the actual data.

[0012] The data processing system according to the present invention includes a memory unit that stores actual data of a three-dimensional chromatogram consisting of a chromatogram and a spectrum obtained by a chromatographic analysis of a sample, and a plurality of peak model functions prepared in advance, and a data processing unit configured to process the actual data of the three-dimensional chromatogram using the plurality of peak model functions and perform a process of separating peaks of a plurality of components that overlap on the chromatogram from each other, wherein the data processing unit is configured to execute an adjustment target peak acquisition step of obtaining a plurality of adjustment target peaks by applying the plurality of peak model functions stored in the memory unit to the chromatogram, and an adjustment target peak adjustment step of setting the adjustment target peak obtained in the adjustment target peak acquisition step to an initial value before adjustment, and repeating adjustment of the adjustment target peak until pseudo data of the three-dimensional chromatogram for the sample obtained by combining the adjustment target peaks after adjustment resembles the actual data.

[0013] That is, in the data processing method and data processing system according to the present invention, first, a peak to be adjusted is obtained using a model-based algorithm, which is estimated as each peak of a plurality of components overlapping each other on a chromatogram. This peak to be adjusted is restricted by a peak model function, so it can be said that it approximates the actual peak to a certain degree, although it does not completely match the actual peak of each component. Such a peak to be adjusted having a certain degree of approximation is obtained, and the peak to be adjusted is used as initial data before adjustment, and adjustment is performed using a model-free algorithm, that is, adjustment is performed by removing the restriction of the peak model function. If an attempt is made to separate each peak of a plurality of components from actual data of a three-dimensional chromatogram using a model-free algorithm from the beginning, the peaks are separated without any information on the peak shape of each component, so it is difficult to obtain an accurate estimation result. On the other hand, if a model-free algorithm is implemented using a peak to be adjusted having a certain degree of approximation as initial data, the peak to be adjusted is fine-tuned to approach the actual data, and the approximation of the estimated data to the actual data is improved. Effect of the Invention

[0014] According to the data processing method and data processing system of the present invention, the peaks to be adjusted of the multiple components overlapping each other on the chromatogram are obtained using the model-based algorithm, and then the peaks to be adjusted obtained by the model-based algorithm are adjusted by the model-free algorithm, so that the peaks to be adjusted are fine-tuned to approach the actual data, and the approximation of the estimated data to the actual data is improved. As a result, even when the peaks of multiple components with extremely high relative concentration ratios in a sample overlap each other on the chromatogram, the quantification of the components with relatively low concentrations can be performed with high accuracy. [Brief description of the drawings]

[0015] [Figure 1] FIG. 1 is a block diagram illustrating an example of a data processing system. [Diagram 2] 2 is a flowchart illustrating a data processing method implemented by the data processing system according to the embodiment. [Diagram 3] 10 is a flowchart showing a specific example of the data processing method. [Figure 4] FIG. 1 shows an example of fitting a peak model function in a peak model-based algorithm, where (A) shows a chromatogram at a wavelength of actual data, and (B) shows the state in which the peak model function is fitted to the same chromatogram. [Diagram 5] This figure shows the peak separation process in the data processing method, where (A) shows a chromatogram at a wavelength with actual data, (B) shows estimated data of the chromatograms of each component separated by a peak model-based algorithm, and (C) shows estimated data of the chromatograms of each component adjusted by matrix decomposition. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an embodiment of a chromatogram data processing method and a data processing system according to the present invention will be described with reference to the drawings.

[0017] FIG. 1 shows an embodiment of a data processing system.

[0018] The data processing system 1 includes an actual data storage unit 2, a peak model storage unit 4, and a data processing unit 6. Analysis data acquired by an analysis device 100 is input into the data processing system 1. The analysis device 100 is configured to perform liquid chromatography analysis on a sample and acquire an absorbance spectrum at regular intervals. In other words, the analysis data input from the analysis device 100 to the data processing system 1 is three-dimensional chromatogram data consisting of a chromatogram and a spectrum. Hereinafter, the three-dimensional chromatogram data input from the analysis device 100 to the analysis system 1 will be referred to as "actual data."

[0019] The actual data storage unit 2 is a storage area for storing the actual data of the three-dimensional chromatogram acquired from the analysis device 100. The actual data storage unit 2 can be realized by a non-volatile flash memory, a hard disk drive, or the like.

[0020] The peak model storage unit 4 stores a plurality of peak model functions prepared in advance. For example, the peak model function may be a model based on an improved EMG function consisting of a combination of a Gaussian function and an exponential function configured to reproduce a peak waveform having a tailing or leading like an actual peak waveform appearing in a chromatogram. The peak model storage unit 4 can be realized by a non-volatile flash memory or a hard disk drive, as with the actual data storage unit 2, but can also be realized by a database provided on a network.

[0021] The data processing unit 6 processes the actual data of the three-dimensional chromatogram stored in the actual data storage unit 2. The processing of the actual data by the data processing unit 6 includes not only a quantitative process for quantifying the concentrations of components contained in a sample from the area values ​​of the peaks on the chromatogram of the actual data, but also a peak separation process for separating peaks of multiple components that overlap each other on the chromatogram of the actual data. The data processing unit 6 is a function realized by executing a program in a computer circuit equipped with a CPU (Central Processing Unit).

[0022] As shown in Fig. 2, the peak separation process by the data processing unit 6 includes a first step (step 101) of creating estimated data of chromatograms and spectra of multiple components whose peaks overlap each other using an algorithm (model-based algorithm) that uses a peak model function, and a second step (step 102) of adjusting the estimated data using an algorithm (model-free algorithm) that does not use a peak model function. The model-based algorithm used in the first step may be a known one, such as one disclosed in Patent Document 1 (International Publication No. WO 2016 / 035167). The peak-free model algorithm used in the second step may also be a known one, such as one that uses matrix decomposition such as NMF (non-negative matrix factorization).

[0023] A more specific procedure for peak separation processing is shown in FIG.

[0024] When the peak separation process is started, the data processing unit 6 first fits a peak model function to the chromatogram of the actual data while adjusting the parameters (height, spread, etc.) of the peak model function, and obtains the fitted peak model function as the adjustment target peak (step 201). For example, if the waveform of a chromatogram at a certain wavelength in the actual data is as shown in FIG. 4(A), the waveform of the chromatogram is approximated by three peak model functions whose parameters have been adjusted, as shown in FIG. 4(B). Each peak model function fitted to approximate the waveform of the chromatogram is the adjustment target peak. This step 201 is the first process using the model-based algorithm.

[0025] The data processing unit 6 synthesizes the peaks to be adjusted obtained in step 201 to generate pseudo data of a three-dimensional chromatogram for the sample (step 202), and calculates the similarity of the pseudo data of the three-dimensional chromatogram to the actual data (step 203). The "similarity" may be any value that numerically represents how similar the pseudo data is to the actual data. Therefore, the method of calculating the similarity is not particularly limited, but the similarity may be calculated, for example, by the sum of the squares of the differences between the numerical values ​​of the pseudo data and the actual data at each point of the three-dimensional chromatogram.

[0026] The data processing unit 6 adjusts the parameters of the adjustment target peak using matrix decomposition so that the calculated similarity is improved, that is, so that the pseudo data becomes closer to the actual data (step 205). Thereafter, the data processing unit 6 synthesizes the adjustment target peaks after the adjustment to generate pseudo data of a three-dimensional chromatogram (step 202), and evaluates the similarity of the generated pseudo data to the actual data (steps 203, 204). In this manner, steps 202 to 205 are repeated, and when the similarity of the pseudo data to the actual data satisfies a predetermined condition, the peak separation process is terminated (step 206: Yes). The predetermined condition may be that the similarity is lower (or higher) than a preset threshold value, or that the similarity of the pseudo data obtained by synthesizing the adjustment target peaks after the adjustment to the actual data converges to a certain value.

[0027] The above steps 202 to 205 are the second step using a model-free algorithm. In this second step, each peak separated in the first step is adjusted without being restricted in shape by the peak model function. As a result, the parts that could not be approximated to the actual data in the first step due to the restrictions of the peak model function are adjusted, and the size and shape of the peak of each component after separation become closer to the actual ones.

[0028] FIG. 5 shows an example of the peak separation state at each step of the peak separation process.

[0029] Fig. 5(A) is a part of the waveform of a chromatogram at a certain wavelength of actual data before peak separation processing is performed. When the first step using the model-based algorithm is performed on the actual data having this chromatogram, two peak model functions are applied to obtain two adjustment target peaks P4 and P5, as shown in Fig. 5(B). Then, when the second step using the model-free algorithm is performed using the obtained adjustment target peaks as the initial data before adjustment, the shapes and sizes of the adjustment target peaks P4 and P5 are adjusted, as shown in Fig. 5(C).

[0030] As described above, when the peaks of two components with an extremely high relative concentration ratio of 100:0.05 are separated using only the model-based algorithm, i.e., by performing only the first step, the peak area of ​​the component with a relatively low concentration can be more than twice the actual peak area. In contrast, the present inventors have confirmed that when the second step is performed using a model-free algorithm with the peak separation result obtained in the first step as the initial data before adjustment, the shape and size of the peaks of the two components are adjusted, and as a result, the peak area of ​​the component with a relatively low concentration approaches the actual peak area.

[0031] The above-described embodiments are merely examples of the data processing method and data processing system according to the present invention. The embodiments of the data processing method and data processing system according to the present invention are as follows.

[0032] In one embodiment of the data processing method according to the present invention, the data processing method uses actual data of a three-dimensional chromatogram consisting of a chromatogram and a spectrum obtained by chromatography analysis of a sample, and separates peaks of multiple components overlapping on the chromatogram from one another, and includes: an adjustment target peak acquisition step of obtaining multiple adjustment target peaks by applying a peak model function prepared in advance to the chromatogram in order to approximate the waveform of the chromatogram to the chromatogram; an estimated data creation step of synthesizing the multiple adjustment target peaks obtained in the adjustment target peak acquisition step to generate estimated data of a chromatogram and a spectrum for each of the multiple components; and an adjustment target peak adjustment step of setting the estimated data created in the estimated data creation step as an initial value before adjustment, and repeating adjustment of the adjustment target peak until pseudo data of the three-dimensional chromatogram for the sample obtained by synthesizing the adjusted adjustment target peaks becomes similar to the actual data.

[0033] In a first aspect of one embodiment of the data processing method, a function that combines a Gaussian function and an exponential function is used as the peak model function. With this aspect, a peak model function that takes into account the tailing and leading of an actual peak can be used, and the estimated shape of the peak after separation can be made closer to the actual one.

[0034] In a second aspect of one embodiment of the data processing method, the adjustment is performed using matrix decomposition in the adjustment target peak adjustment step. This second aspect can be combined with the first aspect.

[0035] As the matrix decomposition, non-negative matrix factorization can be used.

[0036] In a third aspect of one embodiment of the data processing method, in the adjustment target peak adjustment step, a similarity between the pseudo data obtained by combining the adjusted adjustment target peaks and the actual data is calculated, and when the similarity satisfies a preset criterion or when the similarity converges to a certain value, it is determined that the pseudo data is similar to the actual data, and the adjustment is terminated. This third aspect can be combined with the first and / or second aspects.

[0037] In one embodiment of the data processing system according to the present invention, the system includes a memory unit that stores actual data of a three-dimensional chromatogram consisting of a chromatogram and a spectrum obtained by a chromatographic analysis of a sample, and a plurality of peak model functions prepared in advance, and a data processing unit configured to process the actual data of the three-dimensional chromatogram using the plurality of peak model functions and separate peaks of a plurality of components that overlap on the chromatogram from each other, wherein the data processing unit is configured to execute an adjustment target peak acquisition step of obtaining a plurality of adjustment target peaks by applying the plurality of peak model functions stored in the memory unit to the chromatogram, and an adjustment target peak adjustment step of setting the adjustment target peak obtained in the adjustment target peak acquisition step as an initial value before adjustment, and repeating adjustment of the adjustment target peak until pseudo data of the three-dimensional chromatogram for the sample obtained by combining the adjustment target peaks after adjustment resembles the actual data.

[0038] In a first aspect of one embodiment of the data processing system, the data processing unit is configured to use a function that combines a Gaussian function and an exponential function as the peak model function. In this manner, a peak model function that takes into account tailing and leading of an actual peak can be used, and the estimated shape of the separated peak can be made closer to the actual one.

[0039] In a second aspect of one embodiment of the data processing system, the data processing unit performs the adjustment using matrix decomposition in the adjustment target peak adjustment step. This second aspect can be combined with the first aspect.

[0040] As the matrix decomposition, non-negative matrix factorization can be used.

[0041] In a third aspect of one embodiment of the data processing system, the data processing unit is configured to calculate a similarity between the pseudo data obtained by combining the adjusted adjustment-target peaks and the actual data in the adjustment-target peak adjustment step, and terminate the adjustment when the similarity satisfies a preset criterion or when the similarity converges to a certain value. This third aspect can be combined with the first and / or second aspects. [Explanation of symbols]

[0042] 1 Data Processing System 2 Actual data storage section 4 Peak model storage section 6 Data Processing Section 100 Analyzer

Claims

1. 1. A data processing method for separating overlapping peaks of multiple components from each other on a chromatogram obtained by chromatographic analysis of a sample, the method comprising: preparing a peak model function for approximating a waveform of the chromatogram; an adjustment target peak acquisition step of adjusting parameters of the peak model function and applying the function to the chromatogram to obtain a plurality of adjustment target peaks; and an adjustment target peak adjustment step of acquiring a pseudo-chromatogram by using the multiple adjustment target peaks as initial values ​​and readjusting matrix factor parameters of shapes and sizes of the multiple adjustment target peaks using matrix decomposition, and repeating adjustment of the matrix factor parameters of the adjustment target peaks until a similarity between the pseudo-chromatogram and the chromatogram falls below a predetermined threshold or converges to a constant value.

2. The data processing method according to claim 1 , wherein a function combining a Gaussian function and an exponential function is used as the peak model function.

3. The data processing method of claim 1 , wherein the matrix decomposition is a non-negative matrix factorization.

4. a memory unit that stores a chromatogram obtained by a chromatographic analysis of a sample and a peak model function; A data processing unit configured to process the chromatogram using the peak model function and separate peaks of multiple components overlapping on the chromatogram from each other, The data processing unit includes: an adjustment target peak acquisition step of adjusting parameters of the peak model function stored in the storage unit and applying the function to the chromatogram to obtain a plurality of adjustment target peaks; and an adjustment target peak adjustment step of obtaining a pseudo-chromatogram by using the plurality of adjustment target peaks as initial values ​​and readjusting matrix factor parameters of the shapes and sizes of the plurality of adjustment target peaks using matrix decomposition, and repeating adjustment of the matrix factor parameters of the adjustment target peaks until a similarity between the pseudo-chromatogram and the chromatogram falls below a predetermined threshold or converges to a constant value.

5. 5. The data processing system according to claim 4, wherein the data processing unit is configured to use a function that is a combination of a Gaussian function and an exponential function as the peak model function.

6. 5. The data processing system of claim 4, wherein the matrix decomposition is a non-negative matrix factorization.

Images