Information processing device, information processing method, and computer program

Abstract

An information processing device (100) is provided with a model acquisition unit (114), a target data acquisition unit (115), and a classification execution unit (116). The model acquisition unit (114) acquires a trained model (MO) to which comparison data (CD) is input and which outputs a living body classification result, the comparison data (CD) including, with respect to at least one item group pair (P1, P2) that is a pair of two item groups each constituted from at least one item in examination values (ED) for a living body, data indicating the ratio of an examination value (ED) belonging to the other item group to an examination value (ED) belonging to one item group. The target data acquisition unit (115) acquires the comparison data (CD) for a target living body. The classification execution unit (116) executes classification of the target living body by inputting the comparison data (CD) for the target living body to the trained model (MO), and outputs a classification result (RD).

Description

Information processing device, information processing method, and computer program

[0001] The technology disclosed in this specification relates to information processing for classifying living organisms based on test values ​​of the living organisms.

[0002] Techniques for classifying cancer patients based on comprehensive analysis of gene expression have been proposed. For example, Non-Patent Document 1 proposes classifying pancreatic cancer into approximately two to five molecular subtypes with different properties based on the expression patterns of 50 to 700 genes.

[0003] Peter Bailey and 102 others, "Genomic analyses identify molecular subtypes of pancreatic cancer," Nature, UK, Nature Research, February 24, 2016, Vol. 531, No. 7592, pp. 47-52

[0004] The above-mentioned conventional technology classifies pancreatic cancer into multiple types (subtypes) using a technique called unsupervised clustering. This technique determines subtypes based on their relative position in the population, i.e., which subtype they are closest to. Therefore, when applied to a dataset (batch) different from the one used to determine the subtypes, there is a problem in that proper classification may not be possible due to differences in distribution between datasets (batch error).

[0005] Although methods for correcting batch-to-batch errors have been reported, most existing methods that can be applied to a single sample (single-sample classifiers) take an approach of absorbing errors by reducing the amount of information, and the reduced amount of information leads to a problem of reduced accuracy.

[0006] Note that this issue is not limited to the classification of cancer patients based on gene expression levels, but is a common issue in the classification of living organisms based on test values.

[0007] This specification discloses a single sample classifier technique that can solve the above-mentioned problems without reducing the amount of information.

[0008] The technology disclosed in this specification can be realized, for example, in the following forms.

[0009] (1) An information processing device disclosed in this specification includes a model acquisition unit, a target data acquisition unit, and a classification execution unit. The model acquisition unit receives comparison data, including data indicating a ratio of a test value belonging to one item group to a test value belonging to the other item group, for at least one item group pair, each of which is a pair of two item groups consisting of at least one item, for test values ​​of a living organism, and acquires a trained model that outputs a classification result for the living organism. The target data acquisition unit acquires the comparison data for the living organism. The classification execution unit inputs the comparison data for the living organism into the trained model, thereby classifying the living organism and outputting the classification result.

[0010] In this way, this information processing device classifies target organisms using a trained model that inputs comparison data and outputs classification results. The comparison data includes data indicating the ratio of test values ​​belonging to one item group to test values ​​belonging to the other item group for at least one item group pair, which is a pair of two item groups each consisting of at least one item. When selecting items, if items with large differences between the objects to be classified are selected and then the ratio is calculated, a sufficiently large difference will be generated between the objects to be classified, making it possible to relatively reduce the impact of differences in distribution between datasets (batch-to-batch error), thereby improving the accuracy of biological classification.

[0011] (2) In the information processing device, the comparison data may include data indicating the ratio of each test value belonging to the one item group to the average value of all test values ​​belonging to the other item group. This configuration determines the value of the comparison data for each test value, thereby preventing a decrease in the amount of information. This allows comparison data with a larger amount of information to be obtained, effectively improving the accuracy of biological classification.

[0012] (3) In the information processing device, the comparison data may include data for a plurality of the item group pairs. With this configuration, comparison data with a larger amount of information can be obtained, and the accuracy of biological classification can be effectively improved.

[0013] (4) The information processing device may further include a pair setting unit that sets the at least one item group pair based on correlations between a plurality of items. With this configuration, it is possible to set an item group pair that contributes to more appropriate classification, thereby effectively improving the accuracy of biological classification.

[0014] (5) In the information processing device, the test values ​​may be gene expression levels for a plurality of genes. With this configuration, organisms can be classified with high accuracy using comparison data calculated based on the gene expression levels for the plurality of genes.

[0015] (6) In the information processing device, the classification execution unit may be configured to perform the classification for a specific disease. With this configuration, it is possible to improve the accuracy of classification for the specific disease.

[0016] (7) In the information processing device, the specific disease may be cancer. With this configuration, it is possible to improve the accuracy of classification of cancer types.

[0017] (8) In the information processing device, the test value may be a number of amplification cycles of PCR amplification. With this configuration, it is possible to classify living organisms with high accuracy using comparison data calculated based on the number of amplification cycles of PCR amplification.

[0018] (9) In the information processing device, the comparison data used to generate the trained model and the comparison data for the target organism may be based on test values ​​obtained by different testing techniques. This configuration can improve the accuracy of organism classification based on test values ​​obtained by a wide range of testing techniques.

[0019] The technology disclosed in this specification can be realized in various forms, such as an information processing device, an information processing method, a computer program that realizes those methods, a non-transitory recording medium on which that computer program is recorded, etc.

[0020] FIG. 1 is an explanatory diagram showing an outline of the classification model MO in this embodiment; FIG. 2 is an explanatory diagram showing a schematic configuration of the information processing device 100;

[0021] A. Embodiments: A-1. Overview of Classification Model MO: FIG. 1 is an explanatory diagram illustrating the classification model MO in this embodiment. The classification model MO is a trained model for classifying living organisms based on test values ​​of the living organisms. The following describes an example in which the classification model MO is used to classify pancreatic cancer (pancreatic cancer patients) based on gene expression levels obtained by comprehensive gene analysis such as RNA-Seq. Gene expression levels for multiple genes are an example of test values ​​for living organisms. Gene expression levels are data obtained, for example, by RNA-Seq, microarray, RT-PCR, nucleic acid digital counting that detects probes hybridized to target sequences, or the like.

[0022] It is known that there are approximately two to five different molecular subtypes of pancreatic cancer. For example, pancreatic cancer can be classified into multiple types (subtypes) by performing unsupervised clustering using a certain dataset (batch). However, if this classification is applied to a dataset different from the dataset used to set the subtypes, appropriate classification may not be possible due to differences in distribution between the datasets (batch-to-batch error). Therefore, in this embodiment, as described below, comparison data CD identified based on gene expression data ED is used as input to the classification model MO.

[0023] As shown in FIG. 1 , a plurality of gene group pairs (hereinafter simply referred to as "pairs") are set in advance to generate comparison data CD. A gene group pair is a pair of two gene groups (a positive gene group and a negative gene group), each consisting of at least one gene. In the example shown in FIG. 1 , the two gene groups that make up each gene group pair are both made up of multiple genes. For example, the first gene group pair P1 is a pair of a positive gene group consisting of four genes A1, A2, A3, and A4, and a negative gene group consisting of two genes B1 and B2. In this specification, a unit consisting of a single element is also referred to as a "group." A gene group is an example of an item group, and a gene group pair is an example of an item group pair.

[0024] In consideration of the phenomenon of gene expression reversal, the gene group pairs are set as pairs of positive and negative gene groups based on the correlation between multiple genes, as will be described in detail later.

[0025] The comparison data CD for each gene group pair includes data indicating the ratio of the expression level of genes belonging to one gene group (positive gene group or negative gene group) to the expression level of genes belonging to the other gene group (negative gene group or positive gene group). For example, in the example of FIG. 1, the comparison data CD for each gene belonging to one gene group constituting each gene group pair is the value obtained by dividing the expression level of each gene belonging to that one gene group by the average expression level of all genes belonging to the other gene group. Note that, since gene expression levels are generally expressed in logarithmic values, the comparison data CD targeting gene expression levels is the difference between the logarithmic values. For example, for gene A1 belonging to the positive gene group of the first gene group pair P1, the value is the expression level a1 (logarithmic value) of gene A1 minus the average expression level bm (= (b1 + b2) / 2) (also logarithmic value) of all genes (genes B1, B2) belonging to the negative gene group of that pair. Furthermore, for gene B1 belonging to the negative gene group of the pair, the expression level b1 of gene B1 is the value obtained by subtracting the average expression level am (= (a1 + a2 + a3 + a4) / 4) of all genes belonging to the positive gene group of the pair (genes A1, A2, A3, A4).

[0026] In this way, the comparison data CD for each gene group pair is information that indicates the degree to which the expression level of the genes belonging to one gene group is greater than the expression level of the genes belonging to the other gene group. When selecting items, if items that have a large difference between the objects to be classified are selected and then the comparison data CD that calculates the ratio between them is used, a sufficiently large difference will occur between the objects to be classified, and the impact of differences in distribution between data sets (batch error) can be relatively reduced.

[0027] 1, the output of the classification model MO is classification result data RD indicating the classification result. In this embodiment, the classification result is the subtype to which the pancreatic cancer to be classified belongs.

[0028] In this way, for a gene group pair, which is a pair of two gene groups each consisting of at least one gene, comparison data CD including data indicating the ratio of the expression level of genes belonging to one gene group to the expression level of genes belonging to the other gene group is input, and by using a classification model MO, which is a trained model that outputs a classification result for pancreatic cancer, the influence of differences in distribution between datasets (inter-batch error) can be relatively reduced, thereby achieving more accurate classification. Note that the classification method in this embodiment is also referred to as Inverse Pair Boosting (abbreviated as "IPB").

[0029] Among test values ​​for living organisms, the above-mentioned batch-to-batch error is likely to be a major issue for relative values ​​(non-absolute values), and therefore the classification method of this embodiment is particularly useful.

[0030] A-2. Configuration of the information processing device 100: Next, the configuration of the information processing device 100 for creating a classification model MO and performing classification using the classification model MO will be described. Figure 2 is an explanatory diagram showing a schematic configuration of the information processing device 100. The information processing device 100 is configured by, for example, a computer (PC, server, etc.).

[0031] The information processing device 100 includes a control unit 110, a storage unit 120, a display unit 130, an operation input unit 140, and an interface unit 150. These units are connected to each other so as to be able to communicate with each other via a bus 190. The information processing device 100 may also include a speaker as an output means.

[0032] The display unit 130 of the information processing device 100 is configured, for example, by a liquid crystal display or the like, and displays various images and information. The operation input unit 140 is configured, for example, by a keyboard, mouse, buttons, a microphone, a trackpad, or the like, and accepts operations and instructions from an administrator. The display unit 130 may also function as the operation input unit 140 by being equipped with a touch panel. The interface unit 150 is configured, for example, by a LAN interface, a USB interface, or the like, and communicates with other devices via wired or wireless connections.

[0033] The storage unit 120 of the information processing device 100 is configured, for example, with a ROM, RAM, hard disk drive (HDD), etc., and is used to store various programs and data, and as a work area when executing various programs, and as a temporary storage area for data. For example, the storage unit 120 stores a classification program CP, which is a computer program for executing the classification model acquisition process and classification process described below. The classification program CP is provided in a state stored on a computer-readable recording medium (not shown), such as a CD-ROM, DVD-ROM, or USB memory, or is provided in a state retrievable from an external device (a server or other terminal device on a network) via the interface unit 150, and is stored in the storage unit 120 in a state operable on the information processing device 100.

[0034] Furthermore, learning data LD, classification model MO, and classification result data RD are stored in advance or during various processes described below in the storage unit 120 of the information processing device 100. The contents of this information and data will be explained in conjunction with the explanations of the various processes described below.

[0035] The control unit 110 of the information processing device 100 is configured with, for example, a CPU, and controls the operation of the information processing device 100 by executing a computer program read from the storage unit 120. For example, the control unit 110 reads and executes a classification program CP from the storage unit 120, thereby performing various processes described below. In this case, the control unit 110 functions as a raw data acquisition unit 111, a pair setting unit 112, a learning data acquisition unit 113, a model acquisition unit 114, a target data acquisition unit 115, and a classification execution unit 116. The functions of each of these units will be described in conjunction with the explanation of the various processes described below.

[0036] A-3. Classification model acquisition process: Next, the classification model acquisition process executed by the information processing device 100 of this embodiment will be described. FIG. 3 is a flowchart showing the classification model acquisition process in this embodiment. The classification model acquisition process is a process for acquiring the classification model MO described above. In this embodiment, the information processing device 100 acquires the classification model MO by creating the classification model MO by itself using predetermined machine learning. The classification model acquisition process is started in response to a start instruction being input by the user operating the operation input unit 140 of the information processing device 100.

[0037] First, the raw data acquisition unit 111 ( FIG. 2 ) of the information processing device 100 acquires raw data to be used for creating the classification model MO (S110). The raw data is gene expression level data ED (see FIG. 1 ) that indicates the expression levels of each gene for multiple types (subtypes) of pancreatic cancer patients. The raw data is acquired via the interface unit 150 or the operation input unit 140.

[0038] In this embodiment, classification is performed on data that has already been classified using an unsupervised clustering method such as nonnegative matrix factorization (NMF), but when classifying data that has not been subjected to such processing, the target classification may be further divided into smaller subclasses (subclassifications) and each subclassification may be made uniform, thereby improving the prediction accuracy of the target classification itself. For this purpose, the raw data acquisition unit 111 may classify the raw data using an unsupervised clustering method such as nonnegative matrix factorization (NMF) so that each group is uniform.

[0039] Next, the pair setting unit 112 (FIG. 2) of the information processing device 100 selects variant genes between each group as candidate genes (S120). For example, if there are three groups, A, B, and C, variant genes for each of the one-to-one combinations, such as A-B, B-C, and C-A, may be selected as candidate genes. Alternatively, variant genes for each of the remaining pair combinations, such as A-B.C, B-C.A, and C-A.B, may be selected as candidate genes.

[0040] Next, the pair setting unit 112 narrows down the candidate genes selected in S120 (S130). For example, the pair setting unit 112 narrows down the candidate genes using the condition that the absolute value of Log2 fold change (L2FC) is 1 or more, or the condition that the q value is less than 0.01.

[0041] Next, the pair setting unit 112 sets gene group pairs (S140). For example, for each of the one-to-one combinations and one-to-remainder combinations described above, the pair setting unit 112 selects a predetermined number (e.g., approximately 30 to 50) of genes with L2FC > 0 (positive genes) and genes with L2FC < 0 (negative genes) in descending order of the absolute value of L2FC. The pair setting unit 112 then sets gene group pairs, each pair consisting of a positive gene group composed of positive genes and a negative gene group composed of negative genes. In this way, multiple gene group pairs are set as combinations of genes such that when the expression level of one gene increases, the expression level of the other decreases, utilizing the phenomenon of gene expression reversal. Note that the constituent genes may overlap in the gene group pairs set for each of the one-to-one combinations and one-to-remainder combinations.

[0042] Next, the training data acquisition unit 113 ( FIG. 2 ) of the information processing device 100 acquires the above-mentioned comparison data CD for the plurality of set gene group pairs (S150). For example, as shown in FIG. 1 , the training data acquisition unit 113 calculates, as the comparison data CD for the plurality of gene group pairs, a value obtained by subtracting the average value (logarithm) of the expression levels of all genes belonging to one gene group from the expression level (logarithm) of each gene belonging to the other gene group. The training data acquisition unit 113 creates training data LD using the acquired comparison data CD. The training data LD is data in which the comparison data CD is associated with the type (subtype) of pancreatic cancer.

[0043] Note that median correction and missing value imputation may be performed before the process of S150. Furthermore, scaling to make the range of values ​​uniform (for example, scaling to set the median to 0 and 1 SD to 1) may be performed after the process of S150.

[0044] Next, the model acquisition unit 114 ( FIG. 2 ) of the information processing device 100 creates a classification model MO through machine learning using the training data LD (S160). Various known machine learning algorithms can be used for the machine learning used to create the classification model MO. For example, random forest, naive Bayes, support vector machine, etc. may be used to create the classification model MO. An ensemble model (stack model or blended model) combining multiple machine learning algorithms may also be used. The classification model MO may also be created by performing principal component analysis using the comparison data CD described above and performing machine learning using the obtained principal components as inputs. The classification model MO created through machine learning is stored in the storage unit 120 of the information processing device 100. This completes the classification model MO acquisition process ( FIG. 3 ).

[0045] 4, when multiple data sets for training are acquired in S110, the data sets may be merged using a predetermined method (e.g., Combat-seq) (S112) before the processes of S120 to S140 are performed. However, in order to train as a single-sample classifier, such merged data should not be used in S150 and S160, but comparison data CD should be created from the original data and used to create the classification model MO.

[0046] A-4. Diagnosis Processing: Next, the classification processing executed by the information processing device 100 of this embodiment will be described. FIG. 5 is a flowchart showing the classification processing in this embodiment. The classification processing is processing for identifying the type (subtype) of pancreatic cancer of a target patient using the classification model MO. The classification processing is started in response to a start instruction being input by the user operating the operation input unit 140 of the information processing device 100.

[0047] First, the subject data acquisition unit 115 (FIG. 2) of the information processing device 100 acquires comparison data CD for a subject patient (S210). The subject data acquisition unit 115 acquires gene expression level data ED obtained by comprehensive genetic analysis of the subject patient via the interface unit 150, and creates comparison data CD based on the gene expression level data ED. Alternatively, the subject data acquisition unit 115 acquires comparison data CD that has been created in advance based on the gene expression level data ED via the interface unit 150.

[0048] Note that slope correction may be performed before the process of S210. FIG. 6 is an explanatory diagram showing an example of slope correction. Each point in FIG. 6 represents one gene. Although the expression levels of each gene differ between the training data and the data to be evaluated (e.g., test data), if there is no error between the tests, the overall data should correlate along a line SL1 expressed as y = x. If this slope deviates as shown by line SL2 in FIG. 6, a slight deviation will occur in the calculation of IPB. Therefore, in slope correction, the slope is calculated by linear regression, and the data is corrected so that the slope becomes 1.

[0049] Furthermore, median correction and missing value imputation may be performed before the process of S210. Furthermore, scaling may be performed after the process of S210. In the IPB method, the data values ​​for the target patient are the same as those of the training data, so scaling can be performed using statistical values ​​(e.g., median or SD) of the training data.

[0050] Next, the classification execution unit 116 ( FIG. 2 ) of the information processing device 100 performs classification using the comparison data CD for the target patient and the classification model MO (S220). Specifically, the classification execution unit 116 inputs the comparison data CD for the target patient into the classification model MO, and obtains the classification result (information identifying the type of pancreatic cancer) output from the classification model MO. The classification execution unit 116 generates classification result data RD indicating the classification result, and stores it in the memory unit 120 of the information processing device 100.

[0051] Next, the classification execution unit 116 outputs the classification result based on the classification result data RD (S230). For example, the classification execution unit 116 displays the classification result on the display unit 130. This completes the classification process.

[0052] A-5. Example: An example of the above-described classification model MO will be described below. Figure 7 is an explanatory diagram showing classification results in a comparative example. Figure 7 shows the results of principal component analysis performed on three data sets (batches), each showing gene expression levels of multiple pancreatic cancer patients, after correcting inter-batch errors using median correction. In Figure 7, differences in the type (subtype) of pancreatic cancer are represented by different plot shapes, and differences between batches are represented by different hatching. In the results shown in Figure 7, the distributions of the three data sets (batch 1, batch 2, batch 3) are separated from each other, and the differences between batches are relatively large compared to the differences between pancreatic cancer types. Therefore, this method cannot accurately classify pancreatic cancer types.

[0053] FIG. 8 is an explanatory diagram showing classification results in an example. FIG. 8 shows the results of performing principal component analysis on the same three datasets after calculating the above-described comparison data CD according to the method of this embodiment. The method of expressing differences in pancreatic cancer types and batch differences in FIG. 8 is the same as that shown in FIG. 7. In the results shown in FIG. 8, the differences between batches are relatively smaller than the results shown in FIG. 7, making the differences between pancreatic cancer types more clear. This is because the use of the above-described comparison data CD makes the differences between pancreatic cancer types larger relative to the differences between batches. Therefore, the method of this embodiment allows for accurate classification of pancreatic cancer types.

[0054] FIG. 9 is an explanatory diagram showing classification accuracy. FIG. 9 shows the accuracy values ​​for two cases (Case 1 and Case 2) when classification was performed using the method of this embodiment (IPB) and other methods (Switch Box and kTSP+Random Forest). Case 1 is a case in which a classification model MO was created using the data of batch 1 shown in FIGS. 7 and 8 , and batch 2 was classified using this classification model MO. Case 2 is a case in which a classification model MO was created using the same data of batch 1, and batch 3 was classified using this classification model MO. As shown in FIG. 9 , in both Case 1 and Case 2, the method of this embodiment (IPB) was able to classify the type of pancreatic cancer with higher accuracy than the other methods.

[0055] Another example of the IPB method described above is described below. This example demonstrates that differences between testing methods can be relatively reduced by using comparison data CD based on test values ​​obtained using different testing methods. Specifically, in this example, RT-PCR data was mapped to mixed data from RNA-seq and microarray using the IPB method.

[0056] Fig. 10 is an explanatory diagram showing an example of RT-PCR data. As shown in Fig. 11, RT-PCR acquires the number of cycles of amplification (hereinafter also referred to as "Cq value") required for the gene expression level, represented by, for example, fluorescence intensity, to reach a predetermined level Th when the target gene is amplified. Fig. 10 shows the Cq values ​​of each gene belonging to each subtype of pancreatic cancer (subtype I: classical, subtype II: basal-like).

[0057] In RT-PCR, the expression level of the target gene is increased by 2 times the initial value by performing amplification for the Cq value. ＣｑThe Cq value is multiplied by 1. In other words, the larger the Cq value, the smaller the initial value. Therefore, the comparison data CD (IPB value) is obtained by performing the same method as described above using the Cq value multiplied by (-1) (i.e., -Cq). For example, the comparison data CD for gene B belonging to subtype I is calculated by subtracting the average value of -Cq for all genes belonging to subtype II from the -Cq for gene B.

[0058] FIG. 12 is an explanatory diagram showing an example of the results of mapping RT-PCR data to data acquired by another testing method. FIG. 12 shows the results of performing principal component analysis on two data sets, Batch 1, which is RNA-seq data, and Batch 2, which is RT-PCR data, after calculating the above-mentioned comparison data CD according to the IPB method of this embodiment. In FIG. 12, the differences between the two subtypes of pancreatic cancer are represented by different plot shapes, and the differences between batches are represented by different hatching. In the example shown in FIG. 12, the differences between the batches are relatively small, and the differences between pancreatic cancer subtypes are clearly classified.

[0059] FIG. 13 is an explanatory diagram showing another example of the results of mapping RT-PCR data to data acquired by another testing method. Similar to FIG. 12, FIG. 13 shows the results of performing principal component analysis on two data sets, Batch 1, which is RNA-seq data, and Batch 2, which is RT-PCR data, after calculating the above-mentioned comparison data CD according to the IPB method of this embodiment. In FIG. 13, similar to FIG. 12, the differences between the two subtypes of pancreatic cancer are represented by different plot shapes, and the differences between batches are represented by different hatching. In the example shown in FIG. 13, the differences between the batches are relatively small, and the differences between pancreatic cancer subtypes are clearly classified.

[0060] Figure 14 is an explanatory diagram for verifying the degree of integration between batches. For the data in Figure 13, RNA-seq was additionally performed on eight samples from batch 2, which is the RT-PCR data. Figure 14 shows the correspondence between the RNA-seq and RT-PCR results for these eight samples. Specifically, the Y-axis in Figure 14 is a numerical value that serves as an index for classifying subtype I or subtype II on the X-axis (PC1) in Figure 13. In this example, values ​​greater than 0 are generally classified as subtype II, and values ​​less than 0 are classified as subtype I. As shown in Figure 14, the RNA-seq and RT-PCR values ​​for all eight samples are approximately consistent. Note that the difference between the RNA-seq and RT-PCR values ​​is somewhat large for the top sample and the two bottom samples in Figure 14, but even if the numerical difference increases at positions farther from 0, the impact on subtype classification is small. The results in FIG. 14 show that values ​​generated by two completely different modalities, RNA-seq and RT-PCR, can be plotted on the same map by using the IPB method of this embodiment.

[0061] In this way, by using the IPB method of this embodiment, it is possible to relatively reduce the differences between the inspection methods for a plurality of data sets acquired by different inspection methods, and to perform classification with high accuracy.

[0062] Conventionally, the Cq value, which is the result of RT-PCR, has been normalized using the expression level of a housekeeping gene (a gene that is expressed in large amounts and whose expression level is considered to be approximately constant). The IPB method of this embodiment does not use a housekeeping gene, which enables efficiency and cost reduction. Furthermore, because RNA-seq data is used as the reference data, there is no need to collect RT-PCR data for purposes such as setting thresholds as classification standards. A large amount of accumulated RNA-seq data is publicly available on the Internet, and by using this data to construct a model, clinical application of RT-PCR testing can be immediately realized.

[0063] A-6. Effects of this Embodiment: As described above, the information processing device 100 of this embodiment includes a model acquisition unit 114, a target data acquisition unit 115, and a classification execution unit 116. The model acquisition unit 114 receives as input comparison data CD including data indicating the ratio of the expression level of genes belonging to one gene group to the expression level of genes belonging to the other gene group for at least one gene group pair, which is a pair of two gene groups each composed of at least one gene, and acquires a classification model MO, which is a trained model that outputs a classification result for pancreatic cancer. The target data acquisition unit 115 acquires the comparison data CD for the target patient. The classification execution unit 116 inputs the comparison data CD for the target patient into the classification model MO, thereby classifying the target patient's pancreatic cancer and outputting the classification result.

[0064] In this way, the information processing device 100 of this embodiment classifies target patients using the classification model MO, which is a trained model that receives comparison data CD as input and outputs classification results. The comparison data CD includes data indicating the ratio of the expression level of genes belonging to one gene group to the expression level of genes belonging to the other gene group for at least one gene group pair, which is a pair of two gene groups each consisting of at least one gene. Therefore, by using the comparison data CD, the influence of differences in distribution between datasets (batch-to-batch error) can be relatively reduced, thereby improving classification accuracy.

[0065] Furthermore, in the information processing device 100 of this embodiment, the comparison data CD may include data indicating the ratio of the expression level of each gene belonging to one gene group to the average expression level of all genes belonging to the other gene group. With this configuration, the value of the comparison data CD is determined for each gene, thereby preventing a decrease in the amount of information. This allows the comparison data CD to be obtained with a larger amount of information, thereby effectively improving the accuracy of classification.

[0066] Furthermore, in the information processing device 100 of this embodiment, the comparison data CD may include data on a plurality of gene group pairs. With such a configuration, comparison data CD with a larger amount of information can be obtained, and classification accuracy can be effectively improved.

[0067] Furthermore, the information processing device 100 of this embodiment may further include a pair setting unit 112 that sets gene group pairs based on the correlation between multiple genes. With this configuration, it is possible to set gene group pairs that contribute to more appropriate classification, and it is possible to effectively improve the accuracy of classification.

[0068] Furthermore, in the information processing device 100 of this embodiment, the classification execution unit 116 may perform classification for a specific disease. With such a configuration, it is possible to improve the accuracy of classification for the specific disease. Furthermore, the specific disease may be cancer. With such a configuration, it is possible to improve the accuracy of classification for the type of cancer.

[0069] B. Modifications: The technology disclosed in this specification is not limited to the above-described embodiment, and can be modified in various forms without departing from the spirit of the invention. For example, the following modifications are also possible.

[0070] The configuration of the information processing device 100 in the above embodiment is merely an example and can be modified in various ways. Furthermore, the contents of the classification model acquisition process and classification process in the above embodiment are merely an example and can be modified in various ways. For example, in the above embodiment, the information processing device 100 acquires the classification model MO by generating it itself, but the information processing device 100 may also acquire a classification model MO generated by another device.

[0071] The data structure and machine learning algorithm used to create the classification model MO in the above embodiment are merely examples and can be modified in various ways. For example, the comparison data CD used to create the classification model MO may be composed of data for a single gene group pair, or may be composed of data for multiple gene group pairs. Furthermore, in the example of FIG. 1, the comparison data CD for each gene group pair may be calculated by dividing the average expression level of all genes belonging to one gene group by the average expression level of all genes belonging to the other gene group. Furthermore, in the example of FIG. 1, other representative values ​​(e.g., medians) may be used instead of the average value. Furthermore, the testing method for obtaining the data used to create the classification model MO may differ from the testing method for obtaining the input data used to perform classification using the classification model MO. For example, classification may be performed by inputting RT-PCR data into a classification model MO generated based on RNA-seq data.

[0072] In the above embodiment, application to RT-PCR data is described, but the technology disclosed in this specification is not limited to this and can be similarly applied to other test data that detect gene expression levels.

[0073] In the above embodiment, information processing for classifying types of pancreatic cancer based on gene expression levels obtained by comprehensive genetic analysis is exemplified. However, the technology disclosed herein is not limited to pancreatic cancer, and can be similarly applied to classifying types of other cancers and diseases other than cancer. Furthermore, the technology disclosed herein is not limited to gene expression levels, and can also be similarly applied to classifying living organisms based on test values ​​of living organisms. Examples of such test values ​​include blood test values, intestinal flora values, and immunohistochemistry scores.

[0074] In the above embodiments, at least one of the functional units included in the control unit 110 of the information processing device 100 may be included in another device rather than the control unit 110 of the information processing device 100. The classification model acquisition process and the classification process in the above embodiments do not necessarily have to be performed by a single device, and may each be performed by a different device. The steps of the classification model acquisition process in the above embodiments do not necessarily have to be performed by a single device, and may each be performed by a different device. Similarly, the steps of the classification process in the above embodiments do not necessarily have to be performed by a single device, and may each be performed by a different device. In the above embodiments, part of the configuration implemented by hardware may be replaced with software, and conversely, part of the configuration implemented by software may be replaced with hardware.

[0075] 100: Information processing device 110: Control unit 111: Raw data acquisition unit 112: Pair setting unit 113: Learning data acquisition unit 114: Model acquisition unit 115: Target data acquisition unit 116: Classification execution unit 120: Memory unit 130: Display unit 140: Operation input unit 150: Interface unit 190: Bus CD: Comparison data CP: Classification program ED: Gene expression level data LD: Learning data MO: Classification model RD: Classification result data

Claims

1. An information processing device comprising: a model acquisition unit that acquires a trained model that receives comparison data including data indicating a ratio of the test value belonging to one item group to the test value belonging to the other item group for at least one item group pair, which is a pair of two item groups each consisting of at least one item, for test values ​​targeted at a living organism, and outputs a classification result of the living organism; a subject data acquisition unit that acquires the comparison data for a target organism; and a classification execution unit that performs classification of the target organism by inputting the comparison data for the target organism into the trained model, and outputs the classification result.

2. An information processing device according to claim 1, wherein the comparison data includes data indicating a ratio of each of the test values ​​belonging to the one item group to an average value of all of the test values ​​belonging to the other item group.

3. An information processing device according to claim 1 or 2, wherein the comparison data includes data for a plurality of the item group pairs.

4. An information processing device according to claim 1 or 2, further comprising: a pair setting unit that sets the at least one item group pair based on correlations between a plurality of items.

5. An information processing device according to claim 1 or 2, wherein the test value is a gene expression level for a plurality of genes.

6. An information processing device according to claim 5, wherein the classification execution unit performs the classification for a specific disease.

7. An information processing device according to claim 6, wherein the specific disease is cancer.

8. An information processing device according to claim 1 or 2, wherein the test value is the number of amplification cycles of PCR amplification.

9. An information processing device as described in claim 1 or claim 2, wherein the comparison data used to generate the trained model and the comparison data for the target organism are based on test values ​​obtained by different test methods.

10. An information processing device comprising: a model acquisition unit that, for at least one item group pair, which is a pair of two item groups each consisting of at least one item, for test values ​​targeting a living body, receives as input comparison data including data indicating the ratio of the test value belonging to one item group to the test value belonging to the other item group, and generates a trained model by executing learning, which outputs a classification result of the living body.

11. An information processing method comprising the steps of: acquiring a trained model that inputs comparison data including data indicating the ratio of the test value belonging to one item group to the test value belonging to the other item group for at least one item group pair, which is a pair of two item groups each consisting of at least one item, for test values ​​targeted at a living organism, and outputs a classification result for the living organism; acquiring the comparison data for the target living organism; and inputting the comparison data for the target living organism into the trained model, thereby performing classification of the target living organism, and outputting the classification result.

12. An information processing method comprising a step of generating a trained model by executing learning, which inputs comparison data including data indicating the ratio of the test value belonging to one item group to the test value belonging to the other item group for at least one item group pair, which is a pair of two item groups each consisting of at least one item, for test values ​​targeting a living body, and outputs a classification result of the living body.

13. A computer program that causes a computer to execute the following processes: a process of acquiring a trained model that inputs comparison data including data indicating the ratio of the test value belonging to one item group to the test value belonging to the other item group for at least one item group pair, which is a pair of two item groups each consisting of at least one item, for test values ​​of a living organism; a process of acquiring the comparison data for the target organism; and a process of inputting the comparison data for the target organism into the trained model, thereby performing classification of the target organism and outputting the classification result.

14. A computer program that causes a computer to execute a process of generating a trained model by executing learning, in which, for at least one item group pair, which is a pair of two item groups each consisting of at least one item, for test values ​​for a living organism, comparison data including data indicating the ratio of the test value belonging to one item group to the test value belonging to the other item group is input, and a classification result of the living organism is output.

Images