JP2021025953A - Mass analysis data processing method, mass analysis data processing system, and mass analysis data processing program - Google Patents

Abstract

To acquire a large amount of learning data required to construct a highly accurate discrimination model by a small number of mass analysis times.SOLUTION: A mass analysis data analysis method comprise: performing laser ionization of a known sample multiple times in a mass analysis device for ionizing a sample, and acquiring a plurality of pieces of profile data being a spectrum showing a relation between a m/z and strength of an ion generated from the known sample in each laser beam irradiation (Step S11); distributing the plurality of pieces of profile data into a plurality of groups such that each group includes one or more pieces of profile data (Step S12); generating, in each group, a peak list describing the m/z of a peak derived from the known sample and the strength of the peak on the basis of the one or more pieces of profile data included in the group (Step S13); and a generating a discrimination model for discriminating an unknown sample with the peak list and information on a kind of the known sample as learning data.SELECTED DRAWING: Figure 2

Description

The present invention relates to a mass spectrometric data processing method, a mass spectrometric data processing system, and a mass spectrometric data processing program.

The Matrix Assisted Laser Desorption / Ionization (MALDI) method is well known as one of the ionization methods for mass spectrometers. In the MALDI method, in order to analyze a sample that does not easily absorb laser light or a sample that is easily damaged by laser light such as protein, a substance that easily absorbs laser light and is easily ionized is mixed with the sample in advance as a matrix. This is a method of ionizing a sample by irradiating it with a laser beam. In particular, a mass spectrometer using a MALDI ion source (hereinafter referred to as MALDI-MS) can analyze a polymer compound having a large molecular weight without causing too much cleavage, and is also suitable for microanalysis. Therefore, it is widely used in fields such as life science.

Further, in recent years, attempts have been made to discriminate unknown samples by applying machine learning to the mass spectrum obtained by MALDI-MS (see, for example, Patent Document 1). Machine learning is one of the useful methods for finding regularity in a large amount of diverse data and using it to predict, discriminate, or regress data. It is roughly divided into supervised learning. And there is unsupervised learning. For example, when trying to determine the type of a microorganism (for example, species, subspecies, strain, or type) based on the result of analyzing a microorganism with MALDI-MS, a large amount of mass spectrometric data for various microorganisms is previously collected. A discriminant model for discriminating the type of an unknown microorganism is constructed by performing supervised learning using these data as training data (also referred to as teacher data or training data).

JP-A-2018-155522 Japanese Unexamined Patent Publication No. 2010-205460

However, in order to build a highly accurate discrimination model, it is necessary to collect a large amount of training data. For that purpose, it is necessary to perform mass spectrometry many times, which causes a problem that a lot of labor and cost are required.

The present invention has been made in view of the above points, and an object of the present invention is a mass that can obtain a large amount of training data necessary for constructing a highly accurate discrimination model with a small number of mass spectrometrys. It is an object of the present invention to provide an analytical data processing method, a mass spectrometric data processing system, and a mass spectrometric data processing program.

The mass spectrometric data processing method according to the present invention, which has been made to solve the above problems, is
A mass spectrometer that ionizes a sample by laser ionization irradiates a known sample with laser light multiple times, and the relationship between the m / z of ions generated from the known sample and the intensity in each of the multiple laser light irradiations. Acquire multiple profile data, which is a spectrum showing
The plurality of profile data are divided into a plurality of groups so that one or more profile data is included in each group.
For each of the plurality of groups, a peak list describing the m / z of the peak derived from the known sample and the intensity of the peak was generated based on the one or more profile data included in the group.
A discrimination model for discriminating an unknown sample is generated by using the peak list and information on the type of the known sample as learning data.

The mass spectrometric data processing system according to the present invention made to solve the above problems is
The m / z of ions generated from the known sample in each of the multiple laser beam irradiations obtained by performing multiple laser beam irradiations on the known sample in a mass spectrometer that ionizes the sample by laser ionization. A profile data acquisition unit that acquires a plurality of profile data, which is a spectrum showing a relationship with intensity,
A grouping unit that distributes the plurality of profile data into a plurality of groups so that each group includes one or more profile data.
For each of the plurality of groups, a peak list that generates a peak list describing the m / z of the peak derived from the known sample and the intensity of the peak based on the one or more profile data included in the group. With the generator
A discriminant model generator that generates a discriminant model for discriminating an unknown sample by using the peak list and information on the type of the known sample as learning data.
Is provided.

The mass spectrometric data processing program according to the present invention made to solve the above problems causes a computer to function as each part of the mass spectrometric data processing system.

In the mass spectrometry data processing method, the mass spectrometry data processing system, and the mass spectrometry data processing program according to the present invention, the profile data obtained by irradiating one sample with a large number of laser beams is divided into a plurality of groups. , Generate one peak list for each group. This makes it possible to increase the number of peak lists obtained by mass spectrometry for one sample. As a result, it is possible to obtain a large amount of learning data necessary for constructing a highly accurate discrimination model with a small number of mass spectrometrys.

The block diagram which shows the main part structure of the mass spectrometry data processing system which concerns on one Embodiment of this invention. The flowchart which shows the processing procedure of the mass spectrometry data in the same embodiment.

Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a main configuration of a mass spectrometry data processing system 10 according to an embodiment of the present invention.

This system 10 processes mass spectrometric data obtained by analyzing a sample by MALDI-MS (not shown), and includes a learning data generation unit 20, a discrimination model generation unit 30, a discrimination unit 40, and data storage. It includes a unit 50, an input unit 60 including a pointing device such as a mouse and a keyboard, and a display unit 70 including a display device such as a liquid crystal display.

The learning data generation unit 20 performs a predetermined process on the mass spectrometric data obtained by analyzing a known sample (for example, a microorganism to which the strain is known) with MALDI-MS, so that the learning data is used for machine learning. Is to generate. The learning data generation unit 20 includes a profile data acquisition unit 21, a grouping unit 22, and a peak list generation unit 23.

The discrimination model generation unit 30 uses a plurality of training data generated by the training data generation unit 20 to generate a discrimination model for discriminating an unknown sample (for example, a microorganism to which the strain to which it belongs is unknown).

The discrimination unit 40 discriminates the type of the unknown sample (for example, the strain to which the microorganism belongs) by applying the mass spectrometric data obtained by analyzing the unknown sample with MALDI-MS to the discrimination model. .. The discrimination unit 40 includes an unknown sample data acquisition unit 41 and a discrimination execution unit 42.

The substance of the learning data generation unit 20, the discrimination model generation unit 30, and the discrimination unit 40 is a computer (a personal computer or a computer having a higher performance than that), and the dedicated data processing software pre-installed in the computer is used. By operating on a computer, the functions of the above-mentioned parts are realized. The data storage unit 50 may be a storage device built in the computer or directly connected to the computer, or may exist on another computer system that can be accessed from the computer via the Internet or the like, for example. That is, a storage device in cloud computing may be used.

Further, the system 10 according to the present embodiment may have a plurality of computers share the functions of the learning data generation unit 20, the discrimination model generation unit 30, and the discrimination unit 40. Specifically, for example, it is conceivable to assign the functions of the learning data generation unit 20 and the discrimination model generation unit 30 to one computer and the functions of the discrimination unit 40 to another computer.

Subsequently, the characteristics of the processing in the system 10 according to the present embodiment will be described.

Generally, in MALDI-MS, the process of generating ions by laser irradiation → separating and detecting the generated ions is repeatedly executed many times (for example, 120 times) for one sample, and many profiles are used. Data is generated (see Patent Document 2 and the like). The profile data is a data form corresponding to the raw data of the mass spectrometer, and the waveform of the detection signal continuously transmitted from the ion detector provided in the mass spectrometer is shown on the horizontal axis with time. (Or m / z), and the vertical axis represents the ion intensity.

In the conventional data processing method, all the profile data obtained by irradiating one sample with a large number of laser beams as described above are integrated, and then the integration is performed for the convenience of subsequent data processing. The peak included in the waveform of profile data (called integrated profile data) is detected (that is, peak detection processing is performed), and the m / z value indicating the position of the center of gravity (or center position) of each detected peak and the peak It was converted into a list (peak list) showing the area value of. That is, in the conventional data processing method, one peak list is generated as a result of one mass spectrometry for one sample.

On the other hand, in the mass spectrometry data processing method according to the present embodiment, the profile data obtained by irradiating one sample with a large number of laser beams as described above is divided into a plurality of groups, and each group is divided into one group. Generate one peak list. That is, a plurality of peak lists are generated as a result of one mass spectrometry for one sample. This makes it possible to obtain more training data without increasing the number of executions of mass spectrometry.

Hereinafter, the details of such processing will be described with reference to the flowchart of FIG. Here, mass spectrometry is performed on a plurality of known samples (for example, microorganisms whose strains are known) by MALDI-MS in advance, and as a result of mass spectrometry for each of the plurality of known samples, N (N is 2). It is assumed that the profile data (the above integers) is stored in the data storage unit 50 in association with the information on the type of the known sample (for example, the information on the strain of the known microorganism). Hereinafter, the information on the type of the known sample is referred to as a "correct label".

First, when the user performs a predetermined operation on the input unit 60 to specify the mass spectrometric analysis results of the plurality of known samples stored in the data storage unit 50 and instruct the generation of learning data based on these, learning is performed. The data generation unit 20 generates learning data. Specifically, first, the profile data acquisition unit 21 of the learning data generation unit 20 determines the mass spectrometric result of one known sample among the mass spectrometric results of a plurality of known samples designated by the user, that is, N related to the sample. The profile data is acquired from the data storage unit 50 (step S11).

Next, the grouping unit 22 allocates the N profile data to a predetermined M group (M is an integer of N or less) according to a predetermined standard (for example, in the order in which the profile data is generated). (Step S12). At this time, at least one profile data is included in each of the M groups. Also, try to make the number of profile data allocated to each group as even as possible. The number M of the groups may be a value stored in advance on the system 10 side, and may be freely set by the user. Further, the system 10 may automatically determine the number N of profile data or the required discrimination system.

In the ionization of a sample by MALDI, if the same position on the sample is repeatedly irradiated with laser light, ions will gradually disappear, so normally, the laser is located at a plurality of different positions close to each other in the measurement region on the sample. The sample or laser light is moved so that the light irradiates, and profile data is acquired for each of the different positions (measurement points). At this time, the amount of ions generated from each measurement point varies depending on the shade of the sample component in the measurement region. Therefore, in step S12, it is desirable to randomly allocate the N profile data to the M groups. As a result, appropriate learning data can be generated without being affected by the shading of the sample component in the measurement region.

Further, in step S12, a part or all of the N profile data may be allocated to a plurality of groups in duplicate. By doing so, even when the number N of profile data is small or the number M of groups is large, the number of profile data allocated to each group can be increased, so that a decrease in S / N can be prevented. Can be done.

Subsequently, the peak list generation unit 23 generates a peak list for each of the M groups generated in step S12 (step S13). Specifically, the peak list generation unit 23 confirms the number of profile data included in each group, and for a group including a plurality of profile data, the integrated profile data is generated by integrating the plurality of profile data. .. Then, after performing noise removal processing (background removal processing and smoothing processing) on the integrated profile data, peak detection is performed by a predetermined peak detection algorithm. Then, the center of gravity position or center position of the detected peak and the area value of the peak were obtained, and the m / z of the center of gravity position (or center position) of each peak and the area value (corresponding to the intensity) of the peak were described. Generate a peak list. On the other hand, for a group containing only one profile data, noise removal processing (background removal processing and smoothing processing) and peak detection processing are performed on the one profile data without performing the integration processing. Go to generate a peak list. The M peak list thus obtained (that is, the same number as the number of croups) is stored in the data storage unit 50 in association with the correct answer label.

After that, the processing of steps S11 to S13 is performed on all of the plurality of known samples instructed by the user, and M peak lists are generated for each of all the known samples. Here, for the sake of simplification of the explanation, N profile data are acquired for all known samples, the profile data is divided into M groups, and a peak list is generated for each group. , The number N of profile data and the number M of groups (and peak lists) may be different for each sample.

Subsequently, when the user operates the input unit 60 to instruct the generation of the discrimination model using the peak list generated for each of the known samples as learning data, the discrimination model generation unit 30 generates the discrimination model. (Step S14). Specifically, the discrimination model generation unit 30 reads out each M peak list generated for each of the known samples from the data storage unit 50 and the correct answer label associated with each of the peak lists, and learns them. As data, a discrimination model is generated by a predetermined machine learning method. The generated discrimination model is stored in the data storage unit 50. The peak list in the present embodiment is multidimensional data in which m / z of each peak is one dimension, and the discriminant model is, for example, a function of discriminant analysis representing the relationship between multidimensional input and output. ..

The machine learning method used to generate the discrimination model in step S14 is not particularly limited as long as it performs supervised learning, and includes, for example, a support vector machine, a random forest, a neural network, a linear discrimination method, a nonlinear discrimination method, and the like. It is good to do. It is preferable to appropriately select what kind of method is used depending on the type and properties of the data to be analyzed.

After that, an unknown sample to be discriminated (for example, a microorganism whose strain is unknown) is analyzed by MALDI-MS, the obtained peak list is stored in the data storage unit 50, and then the user passes through the input unit 60. Instructs the discrimination of the unknown sample by the discrimination model. In the peak list of the unknown sample, all the plurality of profile data obtained by analyzing the unknown sample with MALDI-MS are integrated, and the integrated profile data is subjected to background removal processing, smoothing processing, and peak detection. It is generated in advance by performing the process. In the discrimination unit 40 that receives the instruction from the user, the unknown sample data acquisition unit 41 reads out the peak list of the unknown sample from the data storage unit 50 (step S15), and the discrimination execution unit 42 reads the unknown in the discrimination model. From the output value obtained by inputting the sample peak list, the type of the unknown sample (for example, the strain to which the unknown microorganism belongs) is determined (step S16).

The discrimination result by the discrimination unit 40 is stored in the data storage unit 50, displayed on the screen of the display unit 70, and presented to the user (step S17).

The mass spectrometry data discrimination system and the mass spectrometry data processing method according to the present embodiment are limited to the generation of a discrimination model for discrimination of microorganisms (discrimination of species, subspecies, strains, types, etc. to which unknown microorganisms belong). However, discrimination of various samples, for example, discrimination of oil type, or discrimination of disease (derived from a biological sample derived from a person suffering from a predetermined disease such as cancer and a person not suffering from the disease) It can be applied to the generation of a discrimination model for discrimination from a biological sample). In addition, the profile data used for generating training data and the peak list of unknown samples to be discriminated in the mass spectrometric data discrimination system and the mass spectrometric data processing method according to the present embodiment are those obtained by analysis by MALDI-MS. Not limited to this, it may be obtained by a mass spectrometer that ionizes a sample by another laser ionization method, for example, a surface assisted laser desorption / ionization method.

The effect of the present invention was verified by the discrimination performance of two types of microorganisms (group A and group B). Here, group A is Escherichia coli, and group B is a microorganism of the genus Achromobacter (Achromobacter. Sp).

First, the sample of group A and the sample of group B were measured four times each by MALDI-MS. At this time, the sample was irradiated with a laser 120 times for each measurement, and 120 profile data were acquired. Then, as an example, a peak list was generated by processing the profile data by the method of the present invention, and a discrimination model using the peak list was generated. Further, as a comparative example, a peak list was generated by processing the profile data by a conventional method, and a discrimination model using the peak list was generated.

Specifically, in the example, when the discrimination model was generated, 120 profile data obtained in one measurement were randomly divided into four groups. Then, 30 profile data included in each group were integrated, and the obtained integrated profile data was subjected to noise removal processing and peak detection processing to generate a single peak list. Using the 32 peak lists (2 groups x 4 measurements x 4 groups) obtained as a result as training data, a discrimination model for discriminating between groups A and B was generated.

On the other hand, in the comparative example, when the discrimination model is generated, all 120 profile data obtained in one measurement are integrated, and the obtained integrated profile data is subjected to noise removal processing and peak detection processing. Generated a single peak list. Using the eight peak lists (2 groups x 4 measurements) obtained as a result as a learning model, a discrimination model for discriminating between groups A and B was generated.

In both the examples and the comparative examples, the statistical analysis software eMSTAT Solution (registered trademark) was used to generate the discrimination model, and SVM (support vector machine) was used as the machine learning algorithm (hereinafter, the same). ..

When the discrimination performance by the discrimination model of the example and the discrimination model of the comparative example was verified, the output result (whether the data is group A or group B) given the test data was 100% correct in both methods. However, the error (estimation error) due to cross-validation was 13% in the model of the comparative example, while it was 0% in the model of the example. The leave-one-out method was used for the cross-validation (the same applies to Examples 2 and 3 described later). That is, one data was extracted from the training data of each group as test data, and machine learning was performed using the remaining data. This was repeated once for all the data until it became test data, and the results were averaged to obtain the estimation error. As a result, it was confirmed that according to the present invention, a discrimination model with higher accuracy than before can be obtained without increasing the number of measurements.

Further, as a further example (Example 2), 120 profile data, which is data for one measurement out of each of the four measurements for the sample of group A and the sample of group B, are combined with 120 groups. Divided into. Then, noise removal processing and peak detection processing were performed on one profile data included in each group to generate a peak list. Using the 240 peak lists (2 groups x 1 measurement x 120 groups) obtained as a result as training data, a discrimination model for discriminating between groups A and B was generated. The reason why the profile data for only one measurement for each group is used for the discrimination model here is to prevent the number of data from becoming too large and the processing from taking a long time.

In addition, as a further example (Example 3), for each of the four measurements of the sample of group A and the sample of group B, 120 profile data obtained in each measurement are randomly divided into two groups. Divided. Then, 60 profile data included in each group were integrated, and a peak list was generated in which the obtained integrated profile data was subjected to noise removal processing and peak detection processing. Using the 16 peak lists (2 groups x 4 measurements x 2 groups) obtained as a result as training data, a discrimination model for discriminating between groups A and B was generated.

When the discrimination performance of the discrimination models obtained in Examples 2 and 3 was verified, it was confirmed that a model with an estimation error of 0% could be generated and that the test data could be 100% correct.

[Various aspects]
It will be understood by those skilled in the art that the above-described exemplary embodiments are specific examples of the following embodiments.

(Clause 1) The mass spectrometric data processing method according to one aspect is
A mass spectrometer that ionizes a sample by laser ionization irradiates a known sample with laser light multiple times, and the relationship between the m / z of ions generated from the known sample and the intensity in each of the multiple laser light irradiations. Acquire multiple profile data, which is a spectrum showing
The plurality of profile data are divided into a plurality of groups so that one or more profile data is included in each group.
For each of the plurality of groups, a peak list describing the m / z of the peak derived from the known sample and the intensity of the peak was generated based on the one or more profile data included in the group.
A discrimination model for discriminating an unknown sample is generated by using the peak list and information on the type of the known sample as learning data.

(Section 2) The mass spectrometric data processing method described in paragraph 1 is
The plurality of profile data may be randomly distributed to the plurality of groups.

(Section 3) The mass spectrometric data processing method according to the first or second paragraph
When distributing the plurality of profile data to the plurality of groups, at least one of the plurality of profile data may be duplicated and distributed to two or more of the plurality of groups.

(Item 4) The mass spectrometric data processing method according to any one of items 1 to 3 is
Further, the unknown sample may be discriminated by applying the peak list generated based on the profile data obtained by mass spectrometry of the unknown sample to the discriminant model.

(Section 5) The mass spectrometry data processing system according to one aspect is
The m / z of ions generated from the known sample in each of the multiple laser beam irradiations obtained by performing multiple laser beam irradiations on the known sample in a mass spectrometer that ionizes the sample by laser ionization. A profile data acquisition unit that acquires a plurality of profile data, which is a spectrum showing a relationship with intensity,
A grouping unit that distributes the plurality of profile data into a plurality of groups so that each group includes one or more profile data.
For each of the plurality of groups, a peak list that generates a peak list describing the m / z of the peak derived from the known sample and the intensity of the peak based on the one or more profile data included in the group. With the generator
A discriminant model generator that generates a discriminant model for discriminating an unknown sample by using the peak list and information on the type of the known sample as learning data.
Is provided.

(Section 6) The mass spectrometry data processing system according to paragraph 5 is
The grouping unit may randomly distribute the plurality of profile data to the plurality of groups.

(Section 7) The mass spectrometric data processing system according to paragraph 5 or 6 is
The grouping unit may duplicately distribute at least one of the plurality of profile data to two or more of the plurality of groups.

(Item 8) The mass spectrometry data processing system according to any one of items 5 to 7 is
A discrimination unit that discriminates the unknown sample by applying the peak list generated based on the profile data obtained by mass spectrometry of the unknown sample to the discrimination model.
May be further provided.

(Section 9) The mass spectrometry data processing program according to one aspect causes a computer to function as each part of the mass spectrometry data processing system according to any one of paragraphs 5 to 8.

According to the mass spectrometric data processing method according to the first paragraph, the mass spectrometric data processing system according to the fifth paragraph, or the mass spectrometric data processing program according to the ninth paragraph, in order to construct a highly accurate discrimination model. It is possible to obtain a large amount of necessary training data with a small number of mass spectrometrys.

Further, according to the mass spectrometry data processing method according to the second item or the mass spectrometry data processing system according to the sixth item, appropriate learning data is obtained without being affected by the shading of the sample component in the measurement region on the sample. Can be generated.

Further, according to the mass spectrometry data processing method described in the third item or the mass spectrometry data processing system described in the seventh item, even if the number of profile data is small or the number of groups is large, the data is allocated to each group. Since the number of profile data to be generated can be increased, it is possible to prevent a decrease in S / N.

10 ... Mass analysis data processing system 20 ... Learning data generation unit 21 ... Profile data acquisition unit 22 ... Grouping unit 23 ... Peak list generation unit 30 ... Discrimination model generation unit 40 ... Discrimination unit 41 ... Unknown sample data acquisition unit 42 ... Discrimination Execution unit 50 ... Data storage unit 60 ... Input unit 70 ... Display unit

Claims (9)

A mass spectrometer that ionizes a sample by laser ionization irradiates a known sample with laser light multiple times, and the relationship between the m / z of ions generated from the known sample and the intensity in each of the multiple laser light irradiations. Acquire multiple profile data, which is a spectrum showing
The plurality of profile data are divided into a plurality of groups so that one or more profile data is included in each group.
For each of the plurality of groups, a peak list describing the m / z of the peak derived from the known sample and the intensity of the peak was generated based on the one or more profile data included in the group.
A mass spectrometric data processing method for generating a discrimination model for discriminating an unknown sample by using the peak list and information on the type of the known sample as learning data. The mass spectrometric data processing method according to claim 1, wherein the plurality of profile data are randomly distributed into the plurality of groups. The invention according to claim 1 or 2, wherein at least one of the plurality of profile data is duplicated and distributed to two or more of the plurality of groups when the plurality of profile data is distributed to the plurality of groups. Mass spectrometric data processing method. Further, according to any one of claims 1 to 3, the peak list generated based on the profile data obtained by mass spectrometry of the unknown sample is applied to the discrimination model to discriminate the unknown sample. Mass spectrometric data processing method. The m / z of ions generated from the known sample in each of the multiple laser beam irradiations obtained by performing multiple laser beam irradiations on the known sample in a mass spectrometer that ionizes the sample by laser ionization. A profile data acquisition unit that acquires a plurality of profile data, which is a spectrum showing a relationship with intensity,
A grouping unit that distributes the plurality of profile data into a plurality of groups so that each group includes one or more profile data.
For each of the plurality of groups, a peak list that generates a peak list describing the m / z of the peak derived from the known sample and the intensity of the peak based on the one or more profile data included in the group. With the generator
A discriminant model generator that generates a discriminant model for discriminating an unknown sample by using the peak list and information on the type of the known sample as learning data.
Mass spectrometric data processing system. The mass spectrometric data processing system according to claim 5, wherein the grouping unit randomly distributes the plurality of profile data to the plurality of groups. The mass spectrometric data processing system according to claim 5 or 6, wherein the grouping unit duplicates at least one of the plurality of profile data into two or more of the plurality of groups. A discrimination unit that discriminates the unknown sample by applying the peak list generated based on the profile data obtained by mass spectrometry of the unknown sample to the discrimination model.
The mass spectrometric data processing system according to any one of claims 5 to 7. A mass spectrometric data processing program that causes a computer to function as each part of the mass spectrometric data processing system according to any one of claims 5 to 8.

Images