Learning data creation method and learning data creation device

Abstract

The present invention is a method for creating training data used for creating an identification device which identifies a plurality of target samples with different properties, wherein measurement data of a plurality of reference samples having known properties is acquired, peak information is extracted by using the first method for the measurement data of the reference sample, a determination is made of whether there are peaks missing by cross-referencing peak information of base samples with the peak information extracted by using the first method, peak information is extracted by using the second method which has a different algorithm and / or parameter from the first method, for the measurement data of the reference sample determined to have a missing peak, a determination is made of whether there are peaks missing by cross-referencing the peak information of the base samples with the peak information extracted by using the second method, and feature quantities are acquired from the measurement data determined to not be missing peaks to create training data.

Description

[Technical field]

[0001] The present invention relates to a technique for creating learning data from measurement data obtained by measuring a sample using an analytical device such as a chromatograph mass spectrometer. [Background technology]

[0002] A high level of quality control is required in each process of food production, processing, distribution, etc., and the demand has been increasing in recent years. Conventionally, evaluation of quality deterioration at food manufacturing and processing sites has generally been based on human subjective evaluation of color, odor, taste, etc. In response to this, attempts have been made to evaluate quality using analytical devices in order to perform more objective and efficient evaluation. For example, Patent Document 1 describes the evaluation of the freshness of fish meat by quantitatively analyzing non-volatile biogenic amines that are generated when fish meat spoils.

[0003] However, food samples contain a wide variety of substances, and there are cases where changes in quality cannot be fully detected by only analyzing the concentration of indicator substances. In addition, in many cases, even foods of the same type and freshness contain different amounts of substances, which can easily lead to erroneous judgments or oversights, especially when the degree of deterioration of the food is low.

[0004] Therefore, it has been attempted to use measurement data obtained by measuring multiple reference samples (food samples, etc.) with known label information (freshness, etc.) indicating the attributes of the target object using a specific analytical device as learning data, and to construct a learning model by machine learning using the learning data. In this method, the measurement data of many reference samples are machine-learned to converge the parameters of the learning model to a level at which the freshness of the target sample can be identified with an accuracy equal to or higher than a predetermined standard, and the classifier is used. The classifier created in this way identifies the freshness of the target sample by determining the correlation of features common to the learning data with the same label information (for example, the position or intensity of the peak of the measurement intensity derived from multiple substances characteristic of reference samples with the same label information (same freshness)). Therefore, it is less likely to cause erroneous judgments or oversights than an analysis based on only a single or specific type of indicator substance. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2017-122677 A [Patent Document 2] JP 2020-165847 A Summary of the Invention [Problem to be solved by the invention]

[0006] In order to create a classifier that can accurately identify the freshness of a target sample, it is necessary to construct a learning model by machine learning using a large amount of learning data from which feature quantities contained in the measurement data of a reference sample have been correctly extracted. For example, when using measurement data (mass chromatograms) obtained by gas chromatography-mass spectrometry, the feature quantities (retention time, mass-to-charge ratio, and measurement intensity) corresponding to each of the peaks derived from each substance contained in the reference sample are extracted to create learning data.

[0007] When extracting peaks from a mass chromatogram (performing peak picking), the user selects a method suitable for the shape of the baseline of the chromatogram from among multiple methods (combinations of a peak extraction algorithm and parameters such as a threshold value for peak height) prepared in advance. Even if the reference samples have the same label information, the amount of substances contained in each reference sample is not the same. In addition, many reference samples may be measured multiple times or using multiple devices. In this case, even if the amount of substances contained in each reference sample is the same, the shape of the baseline and the height of the peak in the measurement data differ slightly for each sample. For these reasons, even if peak picking is performed using the same method and parameters, it may not be possible to extract small peaks contained in the measurement data of some reference samples, and learning data that lacks features corresponding to the peaks that were not extracted will be created. If learning data that lacks features is used, it becomes difficult to converge the parameters of the learning model. In addition, even if the parameters of the learning model converge, the accuracy of classification may decrease when such a learning model is used as a classifier. It is possible to eliminate measurement data from which some peaks are not extracted and create learning data only from measurement data from which all peaks are extracted. However, in that case, the amount of learning data will be reduced by the number of measurement data excluded, which may again make it difficult to converge the parameters of the learning model or reduce the accuracy of the classifier.

[0008] The problem to be solved by the present invention is to provide a technology capable of creating learning data with minimal loss of features when creating learning data by extracting peaks from measurement data obtained by analyzing a reference sample whose sample attributes are known. [Means for solving the problem]

[0009] In order to solve the above problems, the present invention provides a method for creating learning data to be used for creating a classifier for discriminating between a plurality of target samples having different attributes, the method comprising the steps of: acquiring measurement data including a plurality of measurement intensity peaks for a predetermined parameter for each of a plurality of reference samples having known attributes; extracting peak information relating to the plurality of peaks from each of the measurement data of the plurality of reference samples using a first method prepared in advance; determining whether or not a peak is missing by comparing the peak information extracted by the first method with peak information of a previously prepared reference sample having the same attributes; extracting peak information about the plurality of peaks from the measurement data of the reference sample in which the peaks are determined to be missing, using a second method having a different algorithm and / or parameters for extracting peaks from the first method; determining whether or not a peak is missing by comparing the peak information extracted by the second method with the peak information of the reference sample; From the measurement data in which it is determined that there is no missing peak in the peak extracted by the first method or the second method, feature amounts corresponding to each of the plurality of peaks are obtained to create learning data. It is something.

[0010] Another aspect of the present invention, which has been made to solve the above problem, is an apparatus for creating learning data used to create a classifier for discriminating between a plurality of target samples having different attributes, comprising: a measurement data acquisition unit that acquires mass chromatogram data by measurement using a chromatography mass spectrometer for each of a plurality of reference samples whose attributes are known; a reference sample information storage unit that stores peak information included in a mass chromatogram of a reference sample having the same attributes as the reference sample; a method information storage unit storing information on a first method for extracting a peak from the mass chromatogram data and a second method having a different algorithm and / or parameters for extracting a peak from the first method; a first peak extraction unit that extracts peak information regarding the plurality of peaks from each of the mass chromatogram data of the plurality of reference samples by using the first method; a first determination unit that determines the presence or absence of a missing peak by comparing the peak information extracted by the first peak extraction unit with peak information of the reference sample having the same attribute; a second peak extraction unit that extracts peak information regarding the plurality of peaks by using the second method from mass chromatogram data of a reference sample determined by the first determination unit to have a missing peak; a second determination unit that determines the presence or absence of a missing peak by comparing the peak information extracted by the second method with the peak information of the reference sample; a learning data creation unit that creates learning data by acquiring feature amounts corresponding to each of the plurality of peaks from mass chromatogram data determined by the first determination unit or the second determination unit to have no missing peaks; Equipped with. Effect of the Invention

[0011] The above-mentioned attributes include, for example, freshness and place of origin when the reference sample is a food sample, and, for example, the presence or absence of a specific disease when the reference sample is a biological sample.

[0012] In the training data creation method according to the present invention, first, measurement data including a plurality of measurement intensity peaks for a predetermined parameter is obtained for each of a plurality of samples whose attributes are known. In addition, in the training data creation device according to the present invention, mass chromatogram data is obtained as measurement data for each of a plurality of reference samples whose attributes are known by measurement using a chromatographic mass spectrometer. The measurement data may be obtained by actually measuring the samples with an analytical device, or by reading out measurement data obtained in a previous measurement.

[0013] Next, for each of the measurement data of the multiple samples, a peak of the measurement intensity is extracted using a first method prepared in advance. Then, the extracted peaks are compared with peak information of a reference sample having the same attribute prepared in advance to determine whether all peaks have been extracted (whether there are missing peaks). For the peak information of the reference sample, for example, information registered in a library can be used. Alternatively, measurement data of multiple reference samples having the same attribute can be integrated to create integrated measurement data, and information of peaks extracted from the integrated measurement data by the first method can be used as the peak information of the reference sample. The first method and the second method include a combination of an algorithm for extracting peaks and parameters such as a threshold value for peak height. For the algorithm for extracting peaks, a conventional one may be used, or a characteristic algorithm described in the following examples may be used.

[0014] For the measurement data of a sample determined to have missing peaks, a peak of the measured intensity is extracted using a second method having a peak extraction algorithm or parameters different from those of the first method.The peak extracted by the second method is then compared with peak information of a reference sample with the same attributes that has been prepared in advance to determine whether or not the peak is missing.Finally, from the measurement data determined to have no missing peaks extracted by the first or second method, feature amounts corresponding to each of the multiple peaks of the measurement data are obtained to create learning data.

[0015] In the training data creation method and device according to the present invention, training data is created by acquiring features corresponding to the peaks of the measured intensity of the measurement data from the measurement data determined to have no missing peaks, so that training data with reduced missing features can be created. Even if the measurement data does not allow all peaks to be extracted using the first method, training data is created by acquiring features from the measurement data from which all peaks can be extracted using the second method, so that a reduction in the number of training data can be suppressed. [Brief description of the drawings]

[0016] [Figure 1] 1 is a diagram showing the configuration of a main portion of a sample evaluation system including an embodiment of a learning data creation device according to the present invention; [Diagram 2] 3 is a flowchart showing an embodiment of a learning data creation method according to the present invention. [Diagram 3] FIG. 2 is a schematic diagram of mass chromatogram data, which is measurement data acquired in this embodiment. [Figure 4] FIG. 1 is a schematic diagram showing a conventional peak extraction method. [Diagram 5] An example of the results of peak extraction from a mass chromatogram using a conventional peak extraction method. [Figure 6] Another example of the results of extracting peaks from a mass chromatogram using the conventional peak extraction method. [Figure 7] 13 is yet another example of the results of extracting peaks from a mass chromatogram using a conventional peak extraction method. [Figure 8] FIG. 2 is a schematic diagram of baseline definition by the peak extraction method of the present embodiment. [Figure 9] 1 is an example of defining the baseline of a mass chromatogram using the peak extraction method of this embodiment. [Figure 10] 1 shows an example of the results of extracting peaks from a mass chromatogram using the peak extraction method of the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] An embodiment of a method and device for generating learning data according to the present invention will be described below with reference to the drawings.

[0018] The sample evaluation system 1 of the present embodiment is used to estimate the attributes of a target sample. Specifically, the system is used to estimate the freshness or origin of a target sample, such as a food, or to determine whether a subject has a specific disease based on a biological sample.

[0019] 1 shows the configuration of the main parts of a sample evaluation system 1 including the learning data creation device of this embodiment. The sample evaluation system 1 of this embodiment is roughly composed of a gas chromatograph mass spectrometer (GC-MS) 2 and a control / processing unit 3.

[0020] The control and processing unit 3 includes a storage unit 31. The storage unit 31 includes a method information storage unit 311 and a reference sample information storage unit 312. The method information storage unit 311 stores information on a method used to extract peaks from mass chromatogram data obtained by measuring a sample using the gas chromatograph mass spectrometer 2. The method information includes an algorithm for extracting peaks and information on a combination of parameters such as a threshold value for peak height. The algorithm for extracting peaks may include a conventional linking point method, a horizontal method, a new baseline method, or a method using a minimum measurement point (described later) that is characteristic of this embodiment.

[0021] The reference sample information storage unit 312 stores mass chromatogram data of a plurality of reference samples with different attributes for various samples, and peak information (retention time, mass-to-charge ratio, and measurement intensity) for a plurality of peaks appearing in the mass chromatogram. The storage unit 312 also stores measurement conditions and the like when measuring various samples with the gas chromatograph mass spectrometer 2.

[0022] The control and processing unit 3 further includes, as functional blocks, a measurement data acquisition unit 32, a reference sample data creation unit 33, a first peak extraction unit 34, a second peak extraction unit 35, a determination unit 36, a measurement data display unit 37, a learning data creation unit 38, a learning model construction unit 39, and a classifier creation unit 40. The actual substance of the control and processing unit 3 is a personal computer or a more powerful computer called a workstation, and each of the above-mentioned functional blocks is realized by executing a program for the sample evaluation system pre-installed in the computer with the processor of the computer. In addition, an input unit 4 such as a keyboard and a mouse, and a display unit 5 such as a liquid crystal display are connected to the control and processing unit 3.

[0023] Of the components of the control and processing unit 3, the method information storage unit 311 and the reference sample information storage unit 312, and the functional blocks of the measurement data acquisition unit 32, the reference sample data creation unit 33, the first peak extraction unit 34, the second peak extraction unit 35, the judgment unit 36, the measurement data display unit 36, and the learning data creation unit 38 constitute the learning data creation device 10 of this embodiment. Although the learning data creation device 10 is incorporated into a part of the control and processing unit 3 in this embodiment, the learning data creation device 10 may be configured as a device independent of the control and processing unit 3.

[0024] Next, the operation of the sample evaluation system 1 of this embodiment will be described with reference to the flowchart of Fig. 2. In this example, a classifier for estimating the attributes of a target sample is created.

[0025] Prior to creating a classifier, the user sets a reference sample in an autosampler (not shown) connected to the gas chromatograph mass spectrometer 2. The reference sample is of the same type as the target sample and has known attributes. In addition, multiple reference samples are set for each attribute.

[0026] When the user instructs the start of creating a discriminator, the measurement data acquisition unit 32 introduces the reference samples set in the autosampler into the gas chromatograph mass spectrometer 2 in a predetermined order. In the gas chromatograph mass spectrometer 2, the samples are separated into components in the column of the gas chromatograph, then introduced into the mass spectrometer, where each component is ionized by an ionization source such as an electron ionization source, and mass-separated and then detected by an ion detector. The output signals from the ion detector are sequentially sent to the control and processing unit 3, and the measurement data (mass chromatogram data) of each reference sample is stored in the memory unit 31. In this way, mass chromatogram data is acquired for all reference samples (step 1). As shown diagrammatically in FIG. 3, the mass chromatogram data represents the measured intensity of ions with respect to two parameters, time and mass-to-charge ratio, and has a peak of the mass-to-charge ratio of ions generated from a component contained in the sample at the time when the component flows out of the gas chromatogram column (retention time centered on t1).

[0027] When the measurements of all the reference samples are completed and the mass chromatogram data are stored, the reference sample data creation unit 33 groups the reference samples by attribute. Then, the mass chromatogram data of the reference samples having the same attribute is integrated to create integrated mass chromatogram data (step 2). The reference sample data creation unit 33 then activates the first peak extraction unit 34 to create an extracted ion current chromatogram (a chromatogram of the measured intensity of ions of a specific mass-to-charge ratio) from the integrated mass chromatogram data, and extracts peaks using the algorithm and parameters stored in the method information storage unit 311 (step 3).

[0028] Here, the process of extracting peaks from the extracted ion current chromatogram by the first peak extraction unit 34 will be described. Conventionally, for example, the linking point method and the horizontal method have been used as a method for extracting peaks from the extracted ion current chromatogram or the total ion current chromatogram showing the intensity of all ions (FIG. 4). In FIG. 4, the peaks extracted by each method are hatched. In the linking point method, the point where the slope of the waveform of the chromatogram exceeds a predetermined value is set as the peak start point S, and the point where the slope of the waveform falls below the predetermined value is set as the peak end point E. Then, the peak is extracted by defining a baseline by connecting the peak start point S and the peak end point E. In the horizontal method, after defining the peak start point S and the peak end point E in the same manner as above, a horizontal line is drawn through the point with the smaller measured intensity of these two points, and the intersection point with the perpendicular line from the other point is defined as the baseline to extract the peak. Furthermore, when separating a plurality of overlapping peaks, for example, the new baseline method is used. In the new baseline method, a peak start point S and a peak end point E are defined in the same manner as above, and the minimum point located between the two peaks is set as a peak separation point M to separate the peaks.

[0029] For measurement data with little baseline fluctuation, there is no significant difference in the accuracy of peak extraction whether the linking point method or the horizontal method is used. However, for measurement data obtained by measurements using a gas chromatograph, the baseline may increase over time in the latter part of the chromatogram due to a phenomenon called column bleed. Conventionally, peaks are automatically extracted from chromatogram data using preset algorithms such as the linking point method, horizontal method, and new baseline method, but if the set algorithm and parameters are not appropriate, the rise in the baseline due to drift in the latter part of the chromatogram will be extracted as a peak, as shown in Figure 5.

[0030] To deal with this type of drift, a parameter can sometimes be set that takes into account the rise in baseline caused by drift. In Figure 6, this parameter is set to a value of 50, but even with this parameter value set, the rise in baseline caused by drift is still extracted as a peak. On the other hand, in Figure 7, this parameter is set to a value of 100. In Figure 7, the drift is not extracted as a peak, which is appropriate in this respect. However, the baseline slope is too steep in the first half of the chromatogram, making it difficult to obtain the peak height and area correctly.

[0031] Therefore, in this embodiment, the minimum points of the chromatogram waveform are connected to define a baseline and extract peaks. As shown in the upper part of FIG. 8, a chromatogram is formed by connecting a large number of measurement points. In the algorithm of this embodiment, among the measurement points constituting the chromatogram, a measurement point having a smaller measurement intensity than either of the two adjacent measurement points is extracted as a minimum measurement point. Then, the extracted minimum measurement points are linearly interpolated to define a baseline as shown in the lower part of FIG. 8. Note that the method of defining the baseline from the minimum measurement points is not limited to linear interpolation, and an approximation curve connecting these minimum measurement points may be used. Then, a peak is extracted if the height from this baseline exceeds a predetermined threshold value.

[0032] In an extracted ion current chromatogram, peaks do not overlap unless both of the two parameters, retention time and mass-to-charge ratio, are common, so the baseline can be determined by linear interpolation. However, in a chromatogram with one parameter for measurement intensity (such as a total ion current chromatogram), overlapping peaks may appear, and in that case, a minimum point may appear between two peaks. In such a case, linear interpolation of the minimum measurement points including the minimum point between peaks makes it impossible to correctly determine the baseline. In such a case, instead of linearly interpolating the minimum measurement points, for example, an approximation curve for all the minimum measurement points may be obtained to reduce the influence of outliers such as minimum points in the overlapping peaks and correctly determine the baseline. Alternatively, the analyst may delete outliers such as minimum points between peaks and linearly interpolate the other minimum measurement points to determine the baseline.

[0033] Fig. 9 shows the result of defining a baseline for mass chromatogram data having a large drift in the latter half like the chromatograms shown in Figs. 5 to 7, and Fig. 10 shows the result of extracting peaks (circles in the figure indicate peak tops) higher than a predetermined threshold value based on this baseline. As can be seen from these results, even if a large drift exists in the latter half of the chromatogram, the peaks can be correctly extracted by using the method of this embodiment.

[0034] Even if any of the mass chromatogram data of multiple reference samples with the same attribute contains a peak that is difficult to extract, integrating them will compensate for the measured intensities of the other reference samples, making it easier to extract the peak. Therefore, by using the integrated mass chromatogram data, all peaks can be easily extracted. Therefore, in this embodiment, the position information of the multiple peaks extracted from the integrated mass chromatogram data is used as peak information of the reference sample. The reference sample data creation unit 33 stores the peak information of the reference sample for each attribute in the reference sample information storage unit 312.

[0035] Next, the first peak extraction unit 34 extracts peaks from the mass chromatogram data of each reference sample using the algorithm and parameters stored in the method information storage unit 311 in the same manner as described above (step 5).

[0036] When a peak is extracted by the first peak extraction unit 34, the determination unit 36 ​​reads out peak information of a reference sample having the same attributes as the reference sample from the reference sample information storage unit 312. Then, the determination unit 36 ​​compares the peak information of the reference sample with the peak information of the reference sample (step 6).

[0037] When the judgment unit 36 ​​judges that the peak information of the reference sample matches that of the standard sample (i.e., there are no missing peaks) (NO in step 7), it obtains features corresponding to the peaks extracted from the mass chromatogram data of the reference sample to create learning data (step 11). The features extracted here include a pair of retention time and mass-to-charge ratio corresponding to the peak top of each peak. In addition to these, the feature may also include the height or area value of the peak. In other words, something equivalent to a peak list is created as learning data.

[0038] When the judgment unit 36 ​​judges that the peak information of the reference sample does not match the peak information of the standard sample (a part of the peaks to be extracted has not been extracted) (YES in step 7), it displays a screen notifying that a peak is missing from the reference sample. In addition, the measurement data display unit 37 creates an extracted ion current chromatogram corresponding to the mass-to-charge ratio of the missing peak from the mass chromatogram data of the reference sample and displays it on the screen of the display unit 5. The mass chromatogram displayed on the screen has a mark superimposed on a position where the first peak extraction unit 34 did not extract a peak that should have been extracted (i.e., the position of the retention time and mass-to-charge ratio of the peak that is included in the peak information of the standard sample but not included in the peak information of the reference sample). The user can check the waveform of the extracted ion current chromatogram at the position where the mark is displayed and judge whether the second peak extraction unit 35 can extract the peak by changing the algorithm or parameters.

[0039] Furthermore, the determination unit 36 ​​determines whether the number of times the peak extraction process has been performed on the reference sample has reached a predetermined number (predetermined number of times). This number is set to, for example, 5 times. At this stage, only the first peak extraction unit 34 has extracted a peak (the number of times the peak extraction process has been performed is 1), so it is determined that the predetermined number of times has not been reached (NO in step 8).

[0040] If the result of the determination in step 8 is NO, the second peak extraction unit 35 changes the algorithm and / or parameters used in the previous peak extraction (step 9). In this example, the algorithm is not changed, but the threshold value (baseline height) for determining a peak is lowered. Specifically, for example, the threshold value (parameter) used in peak extraction is changed to 90% of the threshold value used in peak extraction by the first peak extraction unit 34 (the threshold value is lowered by 10%).

[0041] Next, the second peak extraction unit 35 uses the changed threshold to extract peaks from the mass chromatogram data of the reference sample again (step 5). Then, it compares again with the peak information of the standard sample (step 6), and if it is determined that there are no missing peaks (NO in step 7), it obtains feature amounts corresponding to each of the extracted peaks to create learning data (step 11). After creating the learning data, it is determined whether processing has been completed for the mass chromatogram data of all the reference samples (step 12). If there is unprocessed data, it performs the processing from step 5 onwards for the mass chromatogram data of the next reference sample in the same manner as above.

[0042] On the other hand, if it is determined again that there is a missing peak (YES in step 7), it is determined whether the number of times the peak extraction process has been performed has reached a predetermined number (step 8), and if the predetermined number has not been reached, the algorithm and / or parameters are changed again (step 9). In this example, the threshold value during the peak extraction process by the first peak extraction unit 34 is set to 100, and the threshold value is lowered by 10 at a time. Alternatively, the threshold value may be lowered by 10% based on the threshold value during the previous peak extraction process.

[0043] If the missing peaks are not resolved even after performing the peak extraction process a specified number of times (YES in step 8), the processing for the mass chromatogram data of that reference sample is terminated, and the processing from step 5 onwards is performed on the mass chromatogram data of the next reference sample in the same manner as described above.

[0044] When the processing of the mass chromatogram data of all the reference samples is completed (YES in step 12), the learning model construction unit 39 constructs a learning model by machine learning using the learning data created in the above procedure (step 13). As a machine learning method, for example, what is called supervised learning can be used. Specifically, in addition to typical machine learning methods such as support vector machines, neural networks, and random forests, multivariate analysis methods such as logistic regression, orthogonal partial least squares, and k-nearest neighbors can be used. When the learning model is constructed by machine learning, the classifier creation unit 40 creates a classifier by the learning model and stores it in the storage unit 31 (step 14).

[0045] Conventionally, when creating learning data, if there is a missing peak extracted from mass chromatogram data of a reference sample whose attributes are known, for example, the mass chromatogram data is excluded, and learning data is created using only the data from which all peaks are extracted. As a result, the number of learning data is reduced by the number of excluded mass chromatogram data, which may make it difficult to converge the parameters of the learning model or reduce the accuracy of the classifier. Alternatively, in Patent Document 2, learning data is created by deleting all information about peaks that do not appear in common in all learning data (are missing in any of the learning data). In this case, learning data is created from which information about peaks that may be useful for determining attributes is deleted, so that it may also be difficult to converge the parameters of the learning model or reduce the accuracy of the classifier.

[0046] In contrast, in this embodiment, even if a peak is missing in the peak extraction process by the first peak extraction unit 34, the second peak extraction unit 35 attempts to extract the peak a predetermined number of times while changing the algorithm or parameters. This makes it possible to prevent a decrease in the number of learning data. In addition, since the learning data is created after confirming that all peaks have been extracted by comparison with the peak information of a reference sample, it is possible to create learning data with no missing features.

[0047] The above embodiment is merely an example and can be appropriately modified in accordance with the spirit of the present invention. The numerical values ​​given in the above embodiment are merely an example and can be appropriately modified according to the characteristics of the target sample and the measurement data.

[0048] In the above embodiment, the user may check the mass chromatogram determined to have a missing peak on the screen of the display unit 5, and only when the user is able to confirm a peak at the position of the missing peak, may the user instruct the second peak extraction unit 35 to extract the peak. With this configuration, learning data can be created efficiently without spending time on processing mass chromatogram data from which it is impossible (or extremely difficult) to extract a peak.

[0049] In the above embodiment, only an algorithm is used to define a baseline from the minimum measurement point and extract a peak, and only the threshold value is changed when the peak is extracted by the second peak extraction unit 35. However, the algorithm for extracting the peak may be changed when the peak is extracted by the second peak extraction unit 35.

[0050] In the above embodiment, mass chromatogram data obtained by measuring a reference sample using the gas chromatograph mass spectrometer 2 was processed, but the same configuration as in this embodiment can be applied to various measurement data including multiple peaks of measurement intensity for a predetermined parameter. Note that the peaks referred to here may also include downward peaks that appear in an absorption spectrum, etc.

[0051] [Aspects] It will be appreciated by those skilled in the art that the exemplary embodiments described above are examples of the following aspects.

[0052] (Section 1) One aspect of the present invention is a method for creating learning data to be used for creating a classifier for discriminating between a plurality of target samples having different attributes. acquiring measurement data including a plurality of measurement intensity peaks for a predetermined parameter for each of a plurality of reference samples having known attributes; extracting peak information relating to the plurality of peaks from each of the measurement data of the plurality of reference samples using a first method prepared in advance; determining whether or not a peak is missing by comparing the peak information extracted by the first method with peak information of a previously prepared reference sample having the same attributes; extracting peak information about the plurality of peaks from the measurement data of the reference sample in which the peaks are determined to be missing, using a second method having a different algorithm and / or parameters for extracting peaks from the first method; determining whether or not a peak is missing by comparing the peak information extracted by the second method with the peak information of the reference sample; From the measurement data in which it is determined that there is no missing peak in the peak extracted by the first method or the second method, feature amounts corresponding to each of the plurality of peaks are obtained to create learning data. It is something.

[0053] (Section 8) Another aspect of the present invention is an apparatus for creating a learning model used in machine learning for creating a learning model constituting a classifier for discriminating between a plurality of target samples having different attributes, the apparatus comprising: a measurement data acquisition unit that acquires mass chromatogram data by measurement using a chromatography mass spectrometer for each of a plurality of reference samples whose attributes are known; a reference sample information storage unit that stores peak information included in a mass chromatogram of a reference sample having the same attributes as the reference sample; a method information storage unit storing information on a first method for extracting a peak from the mass chromatogram data and a second method having a different algorithm and / or parameters for extracting a peak from the first method; a first peak extraction unit that extracts peak information regarding the plurality of peaks from each of the mass chromatogram data of the plurality of reference samples by using the first method; a first determination unit that determines the presence or absence of a missing peak by comparing the peak information extracted by the first peak extraction unit with peak information of the reference sample having the same attribute; a second peak extraction unit that extracts peak information regarding the plurality of peaks by using the second method from mass chromatogram data of a reference sample determined by the first determination unit to have a missing peak; a second determination unit that determines the presence or absence of a missing peak by comparing the peak information extracted by the second method with the peak information of the reference sample; a learning data creation unit that creates learning data by acquiring feature amounts corresponding to each of the plurality of peaks from mass chromatogram data determined by the first determination unit or the second determination unit to have no missing peaks; Equipped with.

[0054] In the method for creating training data in Section 1, first, multiple reference For each sample, measurement data including multiple measurement intensity peaks for a predetermined parameter is obtained. In addition, in the learning data creation device of paragraph 8, for each of multiple reference samples whose attributes are known, mass chromatogram data is obtained as measurement data by measurement using a chromatography mass spectrometer. The measurement data may be obtained by actually measuring the sample with an analysis device, or by reading out measurement data obtained in a previous measurement.

[0055] Next, for each of the measurement data of the multiple samples, a peak of the measured intensity is extracted using a first method prepared in advance.The extracted peaks are then compared with peak information of a reference sample having the same attributes prepared in advance to determine whether all peaks have been extracted (whether or not there are missing peaks).The peak information of the reference sample can be, for example, information registered in a library.

[0056] For the measurement data of a sample determined to have missing peaks, a peak of the measured intensity is extracted using a second method having a peak extraction algorithm or parameters different from those of the first method.The peak extracted by the second method is then compared with peak information of a reference sample with the same attributes that has been prepared in advance to determine whether or not the peak is missing.Finally, from the measurement data determined to have no missing peaks extracted by the first or second method, feature amounts corresponding to each of the multiple peaks of the measurement data are obtained to create learning data.

[0057] In the training data creation method of paragraph 1 and the training data creation device of paragraph 8, training data is created by acquiring features corresponding to peaks of the measured intensity of measurement data from measurement data determined to have no missing peaks, so that training data with reduced missing features can be created. Even in the case of measurement data from which it is not possible to extract all peaks using the first method, training data is created by acquiring features from measurement data from which all peaks can be extracted using the second method, so that a reduction in the amount of training data can be suppressed.

[0058] (Section 2) 2. The learning data creation method according to claim 1, The measurement data determined to have a missing peak in the peak information extracted by the first method or the second method is displayed on a screen.

[0059] (Section 3) 3. The learning data creation method according to claim 2, Peak information is extracted by the second method from the measurement data displayed on the screen that is specified by the user.

[0060] In the learning data creation method of paragraph 2, the user can confirm on the screen the measurement data determined to be missing peaks, thereby confirming whether or not a peak of extractable intensity exists. Also, in the learning data creation method of paragraph 3, the user can specify only the measurement data in which a peak of extractable intensity exists, thereby reducing the processing load related to the extraction of peak information by the second method.

[0061] (Section 4) In the learning data creation method according to any one of claims 1 to 3, Multiple reference Obtaining the measurement data for each of the samples; integrating measurement data of a plurality of reference samples having the same attribute to generate integrated measurement data; As the peak information of the reference sample, peak information regarding a plurality of peaks included in the integrated measurement data is extracted using the first method and used.

[0062] In the method for creating training data described in Section 4, the peak information of the reference sample can be used for samples for which a sufficient number of measurements have not been performed in the past, or for samples for which no measurement data is recorded in a database, such as unknown samples.

[0063] (Section 5) In the learning data creation method according to any one of claims 1 to 4, Peak information on the plurality of peaks is extracted using methods with different algorithms and / or parameters, and a process of determining whether or not the peak is missing is repeated a predetermined number of times until it is determined that the peak is not missing.

[0064] In the learning data creation method of paragraph 5, the process of automatically creating learning data is repeated a predetermined number of times, eliminating the need for the user to decide each time whether or not to perform the process of collecting peak information, thereby reducing the burden on the user.

[0065] (Section 6) Paragraphs 1 to 5 Either In the learning data creation method described in The first method and / or the second method extracts, from among a plurality of measurement points constituting the measurement data, a measurement point whose measurement intensity is lower than both of the measurement intensities of adjacent measurement points as a minimum measurement point; determining a baseline at the plurality of measurement points using the minimum measurement point; At each of the plurality of measurement points, a peak is extracted based on whether a value obtained by subtracting the baseline from the measured intensity at that measurement point exceeds a predetermined threshold value.

[0066] (Section 7) 7. The learning data creation method according to claim 6, Both the first method and the second method use the minimum measurement point to determine a baseline and extract a peak, and the threshold value is different between the first method and the second method.

[0067] In mass chromatogram data obtained by gas chromatography mass spectrometry, the magnitude of the baseline also changes over time. In the learning data creation method of Section 6, the baseline is determined using a minimum measurement point, so it can be suitably used to extract peaks from measurement data in which not only the measurement intensity but also the baseline changes with respect to the parameters. In addition, when extracting peaks from such measurement data, a method in which only the threshold value used to determine whether or not a peak is present can be changed from the first method as in the learning data creation method described in Section 7, and used as the second method. [Explanation of symbols]

[0068] 1. Sample evaluation system 10...Learning data creation device 2. Gas chromatograph mass spectrometer 3. Control and processing section 31...Storage section 311...Method information storage unit 312...Reference sample information storage unit 3 2…Measurement data acquisition section 33…Standard sample data preparation section 34…First peak extraction section 35…Second peak extraction section 36…Judgment section 37…Measurement data display section 38…Learning data creation section 39…Learning model construction section 39... Classifier creation section 4. Input section 5...Display section

Claims

1. A method for creating learning data to be used for creating a classifier for discriminating between a plurality of target samples having different attributes, acquiring measurement data including a plurality of measurement intensity peaks for a predetermined parameter for each of a plurality of reference samples having known attributes; extracting peak information relating to the plurality of peaks from each of the measurement data of the plurality of reference samples using a first method prepared in advance; determining whether or not a peak is missing by comparing the peak information extracted by the first method with peak information of a reference sample having the same attributes that has been prepared in advance; extracting peak information about the plurality of peaks from the measurement data of the reference sample in which the peaks are determined to be missing, using a second method having a different algorithm and / or parameters for extracting peaks from the first method; determining whether or not a peak is missing by comparing the peak information extracted by the second method with the peak information of the reference sample; From the measurement data in which it is determined that there is no missing peak in the peak extracted by the first method or the second method, feature amounts corresponding to each of the plurality of peaks are obtained to create learning data. This is a method for creating training data.

2. 2. The learning data generating method according to claim 1, further comprising the step of displaying on a screen the measurement data determined to have missing peaks in the peak information extracted by the first method or the second method.

3. 3. The learning data generating method according to claim 2, further comprising the step of extracting peak information by the second method from the measurement data displayed on the screen that is designated by a user.

4. acquiring the measurement data for each of a plurality of reference samples having the same attribute; integrating measurement data of a plurality of reference samples having the same attribute to generate integrated measurement data; As the peak information of the reference sample, peak information regarding a plurality of peaks included in the integrated measurement data is extracted using the first method and used.

2. The method for creating learning data according to claim 1,

5. 2. The method for creating learning data according to claim 1, further comprising the steps of: extracting peak information about the plurality of peaks using methods with different algorithms and / or parameters; and repeating a process of determining whether or not the peaks are missing a predetermined number of times until it is determined that the peaks are not missing.

6. The first method and / or the second method extracts, from among a plurality of measurement points constituting the measurement data, a measurement point whose measurement intensity is lower than both of the measurement intensities of adjacent measurement points as a minimum measurement point; determining a baseline at the plurality of measurement points using the minimum measurement point; A peak is extracted based on whether a value obtained by subtracting the baseline from the measured intensity at each of the plurality of measurement points exceeds a predetermined threshold value.

2. The method for creating learning data according to claim 1,

7. The learning data creation method according to claim 6, wherein the first method and the second method both use the minimum measurement points to determine a baseline and extract a peak, and the threshold value is different between the first method and the second method.

8. 1. An apparatus for creating learning data used to create a classifier for discriminating between a plurality of target samples having mutually different attributes, comprising: a measurement data acquisition unit that acquires mass chromatogram data by measurement using a chromatography mass spectrometer for each of a plurality of reference samples whose attributes are known; a reference sample information storage unit that stores peak information included in a mass chromatogram of a reference sample having the same attributes as the reference sample; a method information storage unit storing information on a first method for extracting a peak from the mass chromatogram data and a second method having a different algorithm and / or parameters for extracting a peak from the first method; a first peak extraction unit that extracts peak information regarding the plurality of peaks from each of the mass chromatogram data of the plurality of reference samples by using the first method; a first determination unit that determines the presence or absence of a missing peak by comparing the peak information extracted by the first peak extraction unit with peak information of the reference sample having the same attribute; a second peak extraction unit that extracts peak information regarding the plurality of peaks by using the second method from mass chromatogram data of a reference sample determined by the first determination unit to have a missing peak; a second determination unit that determines whether or not a peak is missing by comparing the peak information extracted by the second method with peak information of the reference sample; a learning data creation unit that creates learning data by acquiring feature amounts corresponding to each of the plurality of peaks from mass chromatogram data determined by the first determination unit or the second determination unit to have no missing peaks; A learning data creation device comprising:

Images