Information processing device, information processing method, and information processing program

Abstract

To provide a drug effect inference model with high reliability.SOLUTION: An information processing device comprises: an arithmetic device which executes predetermined processing; an input unit which accepts related information regarding each of a plurality of events as input; a storage unit which stores a learning model for deriving an inference result of the related information regarding the events; and an output unit which outputs the inference result. The arithmetic device calculates an inner product between a first output value of a hidden layer of the learning model in which a first event of the plurality of event is input and a second output value of the hidden layer of the learning model in which a second event of the plurality of events is input and performs deep learning of the learning model on the basis of the related information regarding the events so that difference between the calculated inner product and correct answer data of the plurality of events becomes small for every combination of the events.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present disclosure relates to an information processing device that learns a drug efficacy prediction model. [Background technology]

[0002] In drug discovery, developing new therapeutic drugs requires a huge amount of time and cost. Therefore, efforts to predict drug efficacy using AI have attracted attention. Predicting drug efficacy using AI makes it possible to confirm the usefulness of a drug for a cohort of interest, but predicting drug efficacy requires processing of a huge amount of complex data. For example, Non-Patent Document 1 discloses a method for presenting features that contribute greatly to prediction using methods such as integrated gradient. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Yi-Ching Tang. Explainable drug sensitivity prediction through cancer pathway enrichment, Scientific Reports, 11: 3128 (2021) Summary of the Invention [Problem to be solved by the invention]

[0004] Generally, when applying drug efficacy prediction AI to clinical applications such as diagnosis, treatment decision-making, and drug discovery, the validity of the results must be judged by experts such as physicians, so the AI ​​requires high levels of interpretability. The technology described in Non-Patent Document 1 predicts drug efficacy based on various types of feature quantities related to sample abnormalities, drug structure, and drug efficacy mechanisms. Furthermore, it calculates an integrated gradient to confirm how and to what extent each feature quantity contributes to the treatment efficacy prediction, such as its effect of improving or reducing the treatment efficacy. However, even if an interpretability index such as an integrated gradient is calculated for various types of feature quantities, the contribution of each feature quantity to the prediction varies, making it difficult to integrate these interpretations and judge validity, and thus it has not been possible to provide a highly reliable drug efficacy prediction model.

[0005] An object of the present invention is to provide a highly reliable drug efficacy prediction model. [Means for solving the problem]

[0006] A representative example of the invention disclosed in the present application is as follows: That is, the invention includes a computing device that executes predetermined processing, an input unit that receives related information related to each of a plurality of events as an input, a memory unit that stores a learning model that derives an inference result of the related information related to the events, and an output unit that outputs the inference result, wherein the computing device calculates an inner product of a first output value of a hidden layer of the learning model to which a first event of the plurality of events is input and a second output value of a hidden layer of the learning model to which a second event of the plurality of events is input, and performs deep learning of the learning model based on the related information related to the events so that, for each combination of the events, the difference between the calculated inner product and correct data of the plurality of events is small. [Effects of the Invention]

[0007] According to one aspect of the present invention, a highly reliable drug efficacy prediction model can be provided. Problems, configurations, and effects other than those described above will become apparent from the following description of the examples. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram illustrating a hardware configuration of an information processing apparatus according to a first embodiment. [Figure 2A] FIG. 2 is a diagram showing an example of the configuration of specimen data according to the first embodiment. [Figure 2B] FIG. 2 is a diagram showing an example of the configuration of drug data according to the first embodiment. [Figure 3] FIG. 10 is a diagram illustrating an example of a setting screen according to the first embodiment. [Figure 4] FIG. 10 is a diagram showing an example of an output screen in the first embodiment. [Figure 5A] 10 is a flowchart of a learning process according to the first embodiment. [Figure 5B] 1 is a flowchart of a test process according to the first embodiment. [Figure 6] FIG. 1 is a diagram illustrating a learning model according to a first embodiment. [Figure 7] FIG. 10 is a diagram illustrating a hardware configuration of an information processing apparatus according to a second embodiment. [Figure 8] FIG. 10 is a diagram illustrating an example of a setting screen according to the second embodiment. [Figure 9] FIG. 10 is a diagram showing an example of an output screen according to the second embodiment. [Figure 10] 10 is a flowchart of a test process according to a second embodiment. [Figure 11] FIG. 11 is a diagram illustrating a hardware configuration of an information processing apparatus according to a third embodiment. [Figure 12] FIG. 10 is a diagram showing an example of the configuration of pathway data in Example 3. [Figure 13] FIG. 11 is a diagram showing an example of a setting screen according to the third embodiment. [Figure 14] FIG. 11 is a diagram showing an example of an output screen according to the third embodiment. [Figure 15] 10 is a flowchart of a test process according to a third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] A preferred embodiment of the information processing device 100 according to the present invention will be described below with reference to the attached screens. In the following description and the attached screens, components having the same functional configuration are designated by the same reference numerals, and redundant description will be omitted.

[0010] Example 1 In Example 1, drug efficacy prediction will be described. That is, the information processing device 100 of Example 1 learns a model in which the dot product of the outputs of each hidden layer learned by deep learning from specimen data 200 and drug data 203 is used as a drug efficacy score. The information processing device 100 of Example 1 also functions as a therapeutic effect prediction device that outputs drug efficacy information of a selected specimen as a drug efficacy prediction result on an output screen 400. A doctor or other expert who is a user can predict drug efficacy using the information processing device 100 and confirm the usefulness of a drug for a cohort of interest.

[0011] FIG. 1 is a diagram illustrating a hardware configuration of an information processing apparatus 100 according to the first embodiment.

[0012] The information processing device 100 is configured as a computer having a processor 101 , a calculation unit 102 , a memory 103 , a storage unit 104 , an output unit 110 , and an input unit 111 .

[0013] The processor 101 is an arithmetic device that executes a program deployed in the memory 103 to realize the functions of the information processing device 100. For example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) can be used as the processor 101. The number of processors 101 is not limited to one, and the functions of the information processing device 100 may be realized by a plurality of processors. Note that part of the processing performed by the processor 101 by executing a program may be executed by another type of arithmetic device (for example, hardware such as an ASIC or an FPGA).

[0014] The memory 103 includes a ROM, which is a non-volatile storage element, and a RAM, which is a volatile storage element. The ROM is a storage device that stores unchanging programs (e.g., BIOS) and the like. The RAM is a high-speed, volatile storage element such as a DRAM (Dynamic Random Access Memory), and temporarily stores programs executed by the processor 101 and data used when the programs are executed. The storage device is also connected to the arithmetic unit.

[0015] The storage unit 104 is configured with a storage device that provides a large-capacity, non-volatile storage area, such as a magnetic storage device (HDD) or a flash memory (SSD). The storage unit 104 also stores data (learning data 105, correct answer data 106, test data 107, inferred data 108, and learning model 109) used by the processor 101 when executing the program, and the program executed by the processor 101. That is, the program is read from the storage unit 104, loaded into the memory 103, and executed by the processor 101 to realize each function of the information processing device 100.

[0016] The training data 105 is data for the training model 109 to learn through machine learning, or data that has already been trained, and includes specimen data 200 and drug data 203. An example of the configuration of the specimen data 200 will be described with reference to FIG. 2A, and an example of the configuration of the drug data 203 will be described with reference to FIG. 2B. The correct answer data 106 is data that is the correct answer for a specific pair of specimen and drug present in the training data 105, and is associated with the training data 105 by a common ID. The correct answer data 106 may be a value measured in the state of the training data 105. For example, the correct answer data 106 is therapeutic efficacy represented by a pharmacological efficacy score associated with the specimen data 200 by a specimen ID 201 and with the drug data 203 by a drug ID 204. The test data 107 is data from which pharmacological efficacy is estimated using the training model 109, and includes the specimen data 200 and drug data 203 in the same format as the training data 105. Details of the training data 105 and the test data 107 will be described later. The inferred data 108 is a drug efficacy score inferred by the learning model 109 from the test data 107 .

[0017] The learning model 109 is a deep learning model for predicting drug efficacy from drug and specimen data, and is a supervised learning model trained using learning data 105 associated with ground truth data 106.

[0018] The output unit 110 is an interface that outputs settings required for program execution and program execution results in a format that can be viewed by the user, and may be configured, for example, as a liquid crystal display. The input unit 111 is an interface that receives input from an operator, and may be configured, for example, as a mouse and keyboard. A touch panel may serve as both the output unit 110 and the input unit 111. Alternatively, a user terminal connected to the information processing device 100 via a network may provide the output unit 110 and the input unit 111. In this case, the information processing device 100 may have a web server function, and the user terminal may access the information processing device 100 using a predetermined protocol (e.g., http). Furthermore, one or both of the output unit 110 and the input unit 111 may be connected to another information processing device and output calculation results to the other information processing device and / or receive data required for calculation from the other information processing device.

[0019] The information processing device 100 may have a network interface device (not shown) that controls communication with other devices in accordance with a predetermined protocol.

[0020] The programs executed by the processor 101 are provided to the information processing device 100 via removable media (CD-ROM, flash memory, etc.) or a network, and are stored in the storage unit 104, which is a non-transitory storage medium. For this reason, the information processing device 100 preferably has an interface for reading data from removable media.

[0021] The information processing device 100 is a computer system configured on a single physical computer or on multiple logically or physically configured computers, and may run on a virtual computer constructed on multiple physical computer resources. For example, multiple programs that realize the functions of the information processing device 100 may each run on separate physical or logical computers, or multiple programs may be combined to run on a single physical or logical computer.

[0022] FIG. 2A is a diagram showing an example of the configuration of sample data 200 included in the training data 105 and test data 107 according to the first embodiment.

[0023] The specimen data 200 included in the training data 105 and the test data 107 is information relating to features associated with specimens, and includes records in which specimen IDs 201 and features 202 are associated with each other.

[0024] FIG. 2B is a diagram showing an example of the structure of the drug data 203 included in the training data 105 and the test data 107 in the first embodiment.

[0025] The drug data 203 included in the training data 105 and the test data 107 is information on features related to drugs, and includes a drug ID 204 and a feature 205. Each drug ID 204 is associated with a feature 205.

[0026] The feature 202 of the specimen data 200 and the feature 205 of the drug data 203 can be, for example, a pathway. As shown in FIG. 12 of Example 3, a pathway is data including one or more edges connecting nodes. In the case of the specimen data 200 and the drug data 203, a pathway is represented by a set of edges in a graph in which nodes represent proteins or genes and edges represent the degree of abnormality occurring at the nodes. The value of the feature 202 of the specimen data 200 can be, for example, gene expression information analyzed using GSEA (Gene Set Enrichment Analysis). Furthermore, when a pathway is used as the feature 202 of the specimen data 200, the feature 202 indicates, for example, the degree of abnormality of a specimen in a specified protein or gene in a specified pathway. Specifically, when a specified specimen is used, if the degree of abnormality occurring in a protein or gene on the pathway is high, the value of the feature 202 will be high, and if the degree of abnormality occurring in a protein or gene on the pathway is low, the value of the feature 202 will be low. The value of the feature 205 of the drug data 203 indicates, for example, the therapeutic effect. Specifically, in the case of a pathway that includes genes that produce functional gene products (e.g., proteins) that increase the therapeutic effect when a specific drug is used, the value of feature 205 will be high, and in the case of a pathway that includes many genes that produce functional gene products (e.g., proteins) that reduce the therapeutic effect, the value of feature 205 will be low.

[0027] The formats of the training data and test data are not limited to those shown in Figures 2A and 2B. For example, the training data and test data may be a single table of data in which the drug data and specimen data are linked in a certain way.

[0028] FIG. 3 is a diagram showing an example of a setting screen 300 output by the information processing apparatus 100 according to the first embodiment.

[0029] The setting screen 300 is displayed on the output unit 110 and is used to set the correct answer data 106 and learning data 105, which are input data for training the learning model 109, to set the test data 107 for predicting drug efficacy using the trained learning model 109, and to set parameters for training and inference.

[0030] The setting screen 300 includes a learning mode button 301 , a learning data file input field 302 , a correct answer data file input field 303 , a test mode button 304 , a test data file input field 305 , a setting button 306 , a setting file input field 307 , an edit button 308 , and a setting button 309 .

[0031] The user switches to learning mode by operating learning mode button 301, and specifies the file of learning data 105 (specimen data 200 and drug data 203) and the file of correct answer data 106 to be used for learning the learning model 109 from the files stored in memory unit 104 in learning data file input field 302 and correct answer data file input field 303, respectively, and can input the specified files into memory 103 by operating setting button 306.

[0032] In addition, the user can switch to test mode by operating test mode button 304, and specify the file of the trained learning model 109 and the test data 107 (specimen data 200 and drug data 203) for evaluating the performance of the learning model 109 from the files stored in memory unit 104 in learning model file input field 314 and test data file input field 305, respectively, and input the specified file into memory 103 by operating setting button 306.

[0033] The user can also specify a setting file that specifies conditions for predicting drug efficacy using the learning model 109 set in the learning model file input field 314, and input the specified file into the memory 103 by operating the setting button 309. The user can also edit the contents of the setting file (e.g., internal parameters of the model, the number of epochs k, and hyperparameters (learning rate and batch size)) using a setting file editor launched by operating the edit button 308. Parameters optimized by prior learning may also be set.

[0034] FIG. 4 is a diagram showing an example of an output screen 400 output by the information processing apparatus 100 according to the first embodiment.

[0035] The output screen 400 is displayed on the output unit 110, and includes a sample selection area 401, a setting button 402, and a drug efficacy information display area 403. The output screen 400 may also include a learning error display area 404.

[0036] The sample selection area 401 includes a sample ID, sample name, and output selection field, and is displayed in table format. The efficacy information display area 403 includes a drug ID, drug name, and efficacy prediction result, and is displayed in table format. Includes efficacy prediction result. When the user selects a sample in the sample selection area 401 and operates the setting button 402, the processor 101 predicts the efficacy using the learning model 109 and outputs efficacy information for the selected sample to the efficacy information display area 403. The efficacy information display area 403 displays drugs in the order in which they contributed to the prediction. The user can confirm the efficacy, drug ID, and drug name predicted for the selected sample by the drugs displayed in the efficacy information display area 403. The learning error display area 404 shows the progress of the loss function over the number of epochs.

[0037] 5A is a flowchart of the learning process executed by the information processing device 100 of Example 1. In the learning process, a learning model 109 is generated using learning data 105 (specimen data 200, drug data 203) and supervised data 106.

[0038] In step S501, the processor 101 executes the process of The processor 101 receives input of the specified learning data 105, correct answer data 106, and internal parameter settings. The processor 101 also sets the number of epochs k=1.

[0039] In step S502, the processor 101 inputs the specimen data 200 and drug data 203 included in the training data 105 into the training model 109 and outputs the hidden layer values.

[0040] In step S503, the processor 101 calculates the dot product of the output value 602 of the hidden layer of the specimen and the output value 603 of the hidden layer of the drug as a drug efficacy score. As shown in FIG. 6, the learning model 109 has a deep learning model 600 to which specimen data 200 is input and a deep learning model 601 to which drug data 203 is input, and calculates the dot product of the output value 602 of the hidden layer of the deep learning model 600 and the output value 603 of the hidden layer of the deep learning model 601. The hidden layer of the deep learning model 600 may represent the degree of genetic abnormality in the specimen, and the layer of the deep learning model 601 may represent the degree of therapeutic effect of the drug. Furthermore, the deep learning models 600 and 601 have the same number of layers from the input layer to the hidden layer. Furthermore, the output of the hidden layer of the deep learning model 600 and the output of the hidden layer of the deep learning model 601 have the same number of dimensions. By making the number of dimensions equal, it becomes possible to calculate the similarity between multiple events using the inner product of the outputs of the hidden layer. In deep learning model 600, by calculating drug efficacy using the inner product of the outputs of the hidden layer, it becomes possible to analyze the relationship between multiple events and predetermined related information such as pathways, even if multiple data types are different.

[0041] In step S504, the processor 101 calculates the value of the loss function based on the efficacy score and the correct answer data 106.

[0042] In step S505, the processor 101 determines whether a learning termination condition is met. For example, the efficacy score calculated in step S504 is compared with the correct answer data 106. If the value of the loss function is equal to or less than a threshold for a predetermined number of consecutive times, the desired learning model has been obtained, and learning may be terminated. Also, if the number of epochs k is equal to or greater than a predetermined value set as the learning termination condition, learning may be terminated.

[0043] In step S506, if the learning termination condition is not met, the processor 101 updates the deep learning models 600 and 601 to minimize the value of the loss function calculated in step S505, adds 1 to the number of epochs k, returns to step S502, and continues processing the next record of the learning data 105.

[0044] In step S507, if the learning termination condition is met, the processor 101 stores the learning model 109 including the updated deep learning models 600 and 601 in the memory unit 104.

[0045] 5B is a flowchart of the test process executed by the information processing device 100 of Example 1. In the test process, the performance of the generated learning model 109 is verified using the test data 107 (specimen data 200, drug data 203) and the correct answer data 106.

[0046] In step S509, the processor 101 receives input of the learning model 109 selected by the user on the output screen 400, the specimen test data 107, the correct answer data 106, and the internal parameter settings.

[0047] In step S510, the processor 101 inputs the specimen data 200 and drug data 203 contained in the test data 107 into the learning model 109 and outputs the values ​​of the hidden layer.

[0048] In step S511, the processor 101 calculates the inner product of the output value 602 of the hidden layer of the specimen and the output value 603 of the hidden layer of the drug as the efficacy score.

[0049] In step S512, the processor 101 calculates the value of the loss function based on the efficacy score and the correct answer data 106.

[0050] In step S513, the processor 101 outputs display data for displaying on the output unit 110 the efficacy of each drug and the learning error due to the loss function.

[0051] FIG. 6 is a diagram showing the learning model 109 according to the first embodiment.

[0052] As described above, the learning model 109 has a deep learning model 600 to which the specimen data 200 is input and a deep learning model 601 to which the drug data 203 is input, and is trained by inputting the drug data 203 into the deep learning model 600 and the specimen data 200 into the deep learning model 601, so that the dot product of the outputs of the hidden layers becomes the correct data 106.

[0053] The output value 602 of the hidden layer of the specimen obtained by inputting specimen data 200 into deep learning model 600 represents the degree of genetic abnormality at each bit in the hidden layer. Also, the output value 603 of the hidden layer of the drug obtained by inputting drug data 203 into deep learning model 601 represents the degree of therapeutic effect of the drug at each bit in the hidden layer.

[0054] The number of layers in deep learning model 600 is the same as the number of layers in deep learning model 601, and the number of dimensions in deep learning model 600 is the same as the number of dimensions in deep learning model 601.

[0055] In this way, the dot product of the outputs of the hidden layers of the two deep learning models 600 and 601 is used as the efficacy score and is trained to be close to the correct data 106, so that the efficacy scores of features that are strongly related to abnormalities and to the therapeutic effects of drugs become high.

[0056] As described above, the information processing device 100 of Example 1 includes an arithmetic unit (processor 101) that executes predetermined processing, an input unit 111 that receives as input related information (specimen data 200, drug data 203) related to each of a plurality of events, a memory unit 104 that stores a learning model 109 that derives inference results for the related information related to the events, and an output unit 110 that outputs the inference results. The arithmetic unit calculates an inner product of a first output value 602 of a hidden layer of the learning model 109 to which a first event (specimen data 200) of the plurality of events is input, and a second output value 603 of a hidden layer of the learning model 109 to which a second event (drug data 203) of the plurality of events is input. For each combination of the events (specimen, drug), the learning model is deep-trained based on the related information related to the events so that the difference between the calculated inner product and the correct answer data 106 of the plurality of events is small. This makes it possible to provide a highly reliable pharmacological efficacy prediction model, and the user can confirm the inferred pharmacological efficacy and drug for each of the specimen data 200.

[0057] In addition, conventionally, based on trial results using public data, the output of the final layer of the model for each sample-drug pair was reduced to two dimensions using tSNE, the latent space was visualized, and the clusters formed for each drug were observed to identify overfitting to drug differences. This means that drug differences significantly contribute to the prediction of efficacy, and predictions are not made based on the relationship between the expected drug's inhibitory target and the sample abnormality. However, in the predictions made by the learning model 109 of Example 1, data related to the sample and drug are independently trained, and the difference between the dot product of the output of the final layer and the correct data 106 is minimized. This prevents overfitting to drug and sample distinctions, while generating a learning model 109 incorporating a prediction mechanism that is expected to be effective when the drug inhibits the sample abnormality.

[0058] Although the first embodiment describes the generation of a learning model, the input unit 111 may newly receive specimen data and drug data as input, and the processor 101 may use the specimen data and drug data to calculate the therapeutic efficacy when the drug data is used for the specimen data, using a learning model trained with the specimen data and drug data. This allows the therapeutic efficacy of the drug for the specimen data to be calculated with high accuracy. The calculated therapeutic efficacy may also be output by the output unit 110.

[0059] <Example 2> In Example 2, the interpretability is improved by calculating the contribution to prediction of each feature of the drug data 203 and the specimen data 200 using the device described in Example 1. In Example 2, differences from Example 1 described above will be mainly explained, and the same configurations and processes as those in Example 1 will be assigned the same reference numerals, and explanations thereof will be omitted.

[0060] FIG. 7 is a diagram illustrating a hardware configuration of an information processing apparatus 100 according to the second embodiment.

[0061] The information processing apparatus 100 of the second embodiment is configured as a computer having a processor 101, a calculation unit 102, a memory 103, a storage unit 104, an output unit 110, and an input unit 111.

[0062] The processor 101, the calculation unit 102, the memory 103, the output unit 110, and the input unit 111 are the same as those in the first embodiment.

[0063] The storage unit 104 is configured with a storage device that provides a large-capacity, non-volatile storage area, such as a magnetic storage device (HDD) or a flash memory (SSD). The storage unit 104 also stores data (learning data 105, correct answer data 106, test data 107, inferred data 108, learning model 109, and contribution matrix group 112) used by the processor 101 when executing the program, and the program executed by the processor 101. That is, the program is read from the storage unit 104, loaded into the memory 103, and executed by the processor 101 to realize each function of the information processing device 100.

[0064] The contribution matrix group 112 indicates the contributions of the specimen data 200 and the drug data 203, and its configuration will be described with reference to FIG.

[0065] FIG. 8 is a diagram showing an example of a setting screen 300 output by the information processing apparatus 100 according to the second embodiment.

[0066] The setting screen 300 of Example 2 includes a learning mode button 301, a learning data file input field 302, a correct answer data file input field 303, a test mode button 304, a learning model file input field 314, a test data file input field 305, a setting button 306, a setting file input field 307, an edit button 308, a setting button 309, a contribution calculation selection field 310, a drawing pair input field 311, and a setting button 312.

[0067] The learning mode button 301, learning data file input field 302, correct answer data file input field 303, test mode button 304, test data file input field 305, settings button 306, settings file input field 307, edit button 308, and settings button 309 are the same as those in the first embodiment described above.

[0068] The user selects whether to calculate and output the contribution in a contribution calculation selection field 310, specifies a file representing the drug-analyte pair for which the contribution is to be output in a drawing pair input field 311, and operates a setting button 312 to input the specified file into memory 103, and the contribution can be displayed on output screen 400. An input field for setting the number of features that contribute greatly to prediction to be displayed may also be provided.

[0069] FIG. 9 is a diagram showing an example of an output screen 400 output by the information processing apparatus 100 according to the second embodiment.

[0070] The output screen 400 of Example 2 is displayed on the output unit 110 and includes a sample selection area 401 , a setting button 402 , a drug efficacy information display area 403 , a learning error display area 404 , a contribution visualization area 405 , and a detail display area 406 .

[0071] The specimen selection area 401, setting button 402, efficacy information display area 403, and learning error display area 404 are the same as those in the first embodiment described above.

[0072] The contribution visualization area 405 outputs the features that have a large contribution to drug efficacy in each of the drug data 203 and the specimen data 200, and the contribution of those features. Furthermore, when the user operates the detailed display button, four diagrams are output in the detailed display area 406. IGdrug is the contribution of the drug feature to drug efficacy, IGcell is the contribution of the specimen feature to drug efficacy, IGmul is the element product of IGdrug and IGcell, and IGsum is the dimensional sum of IGmul. IGdrug, IGcell, and IGmul each represent features (P1 to P5) vertically and each neuron in the hidden layer horizontally.

[0073] The learning process of the information processing apparatus 100 in the second embodiment is the same as that in the first embodiment described above.

[0074] FIG. 10 is a flowchart of the test process executed by the information processing apparatus 100 according to the second embodiment.

[0075] Steps S509 to S512 of the test process are the same as those in the first embodiment described above.

[0076] In step S514, the processor 101 calculates an Integrated Gradient to calculate the contribution of neurons in the hidden layer to drug efficacy. For example, the contribution matrix IGdrug of the drug data 203 is calculated using Equation 1. In Equation 1, i is a subscript related to the feature, j is a subscript of the neuron in the final hidden layer, x is a feature vector, x' is the baseline, F(x) is the output of the final hidden layer, and m is an integer. The contribution matrix IGcell of the specimen data 200 is also calculated in a similar manner using Equation 1. This allows the effect of the drug data 203 on drug efficacy to be calculated as a contribution.

[0077]

number

[0078] The contribution matrix IGcell of the specimen data 200 also has a similar effect, and the effect on the efficacy of the drug in the specimen data 200 can be calculated.

[0079]

number

[0080] In S515, the processor 101 calculates the element product IGmul of the contribution matrices of the drug data 203 and the specimen data 200 using Equation 3.

[0081]

number

[0082] In S516, the processor 101 calculates the contribution IGsum for each feature amount from IGmul using Equation 4.

[0083]

number

[0084] In S517, the processor 101 sorts the IGsums by size and outputs a predetermined number of the top ones.

[0085] In the above-mentioned formula, the integrated gradient of the drug data 203 satisfies formula 5.

[0086]

number

[0087] Similarly, the integrated gradient of the specimen data 200 satisfies Equation 6.

[0088]

number

[0089] Based on the relationship between Formulas 5 and 6, the relationship between the contribution IGsum for each feature and the drug score is expressed as Formula 7. A feature with a larger contribution IGsum can be considered to be a feature that contributes more to improving drug efficacy.

[0090]

number

[0091] In S513, the processor 101 outputs to the output screen 400 the feature amounts with the greatest contributions and the contribution rates of the feature amounts.

[0092] As described above, the information processing device 100 according to the second embodiment of the present invention calculates the contribution of each feature amount, thereby making it possible to know the genetic abnormalities and drug actions that contribute to treatment, thereby increasing the reliability of a drug efficacy prediction model in clinical applications.

[0093] Example 3 In Example 3, in order to utilize it in actual clinical applications, pathways, which are features that contribute greatly to prediction, are illustrated to support decision-making. In Example 3, differences from Examples 1 and 2 described above will be mainly explained, and the same configurations and processes as in Examples 1 and 2 will be assigned the same reference numerals, and their explanation will be omitted.

[0094] FIG. 11 is a diagram illustrating a hardware configuration of an information processing apparatus 100 according to the third embodiment.

[0095] The information processing apparatus 100 of the third embodiment is configured as a computer having a processor 101, a calculation unit 102, a memory 103, a storage unit 104, an output unit 110, and an input unit 111.

[0096] The processor 101, the calculation unit 102, the memory 103, the output unit 110, and the input unit 111 are the same as those in the first embodiment.

[0097] The storage unit 104 is configured with a storage device that provides a large-capacity, non-volatile storage area, such as a magnetic storage device (HDD) or a flash memory (SSD). The storage unit 104 also stores data (learning data 105, correct answer data 106, test data 107, inferred data 108, learning model 109, contribution matrix group 112, and pathway data 113) used by the processor 101 when executing the program, and the program executed by the processor 101. That is, the program is read from the storage unit 104, loaded into the memory 103, and executed by the processor 101 to realize each function of the information processing device 100.

[0098] A pathway defined by the pathway data 113 is composed of one or more nodes that connect edges. An example of the configuration of the pathway data 113 will be described with reference to FIG.

[0099] FIG. 12 is a diagram showing an example of the configuration of the pathway data 113 according to the third embodiment.

[0100] The pathway data 113 includes an edge ID 207, a node ID 208, a node ID 209, and a pathway ID 210. The edge ID 207 is identification information for the edge. The node IDs 208 and 209 are identification information for the nodes located at the ends of the edge. The pathway ID 210 is identification information for the pathway that includes the edge.

[0101] FIG. 13 is a diagram showing an example of a setting screen 300 output by the information processing apparatus 100 according to the third embodiment.

[0102] The setting screen 300 of Example 3 includes a learning mode button 301, a learning data file input field 302, a correct answer data file input field 303, a test mode button 304, a learning model file input field 314, a test data file input field 305, a setting button 306, a setting file input field 307, an edit button 308, a setting button 309, a contribution calculation selection field 310, a drawing pair input field 311, a setting button 312, and an illustration selection field 313.

[0103] The learning mode button 301, learning data file input field 302, correct answer data file input field 303, test mode button 304, test data file input field 305, setting button 306, setting file input field 307, edit button 308, setting button 309, contribution calculation selection field 310, drawing pair input field 311, and setting button 312 are the same as those in the first and second embodiments described above.

[0104] The user selects whether to calculate and output the contribution in a contribution calculation selection field 310, specifies a file representing the drug-analyte pair for which the contribution is to be output in a drawing pair input field 311, selects whether to illustrate the pathway in a drawing selection field 313, and operates a setting button 312 to input the specified file into memory 103, and the contribution can be displayed on the output screen 400. An input field for setting the number of features that contribute greatly to the prediction to be displayed may also be provided.

[0105] FIG. 14 is a diagram showing an example of an output screen 400 output by the information processing apparatus 100 according to the third embodiment.

[0106] The output screen 400 of Example 3 is displayed on the output unit 110 and includes a sample selection area 401, a setting button 402, a drug efficacy information display area 403, a learning error display area 404, a contribution visualization area 405, a detailed display area 406, and a pathway display area 407.

[0107] The specimen selection area 401, setting button 402, efficacy information display area 403, learning error display area 404, contribution visualization area 405, and detailed display area 406 are the same as those in the first and second embodiments described above.

[0108] The pathway display area 407 illustrates pathways that are features that contribute significantly to the prediction. In the illustrated example, the solid lines displayed above the edges are pathways (features) that contributed to the efficacy score in the drug data 203, and the dotted lines displayed below the edges are pathways (features) that contributed to the efficacy score in the sample data 200. The thickness of these solid and dotted lines indicates the magnitude of the contribution. Note that the display mode of pathways that contribute significantly is not limited to the mode described above.

[0109] The learning process of the information processing apparatus 100 according to the third embodiment is the same as that according to the first embodiment.

[0110] FIG. 15 is a flowchart of the test process executed by the information processing apparatus 100 according to the third embodiment.

[0111] Steps S509 to S517 of the test process are the same as those in the first and second embodiments described above.

[0112] In step S519, the processor 101 draws a pathway diagram based on the pathway data 113 and the contribution of each pathway.

[0113] As described above, the information processing device 100 according to the third embodiment of the present invention illustrates features (pathways) that contribute greatly to prediction, allowing intuitive understanding and comparison of their contributions, thereby enabling rational and effective clinical decisions.

[0114] The above describes an example in which the feature is a pathway with specimens and drugs as events, and the inference target is drug efficacy, but the present invention can be applied to other feature values ​​and inference targets. Also, the above describes an example in which the input data is specimen data and drug data, but the present invention can be applied to other input data. For example, the input data may be the number of roads and the number of accidents, the feature may be roads, and the inference target may be the degree of congestion. Furthermore, the number of input data may be three or more, rather than two. For example, the three input data may be the drug and specimen in the above-described example, and the relevance to the disease in each pathway.

[0115] The present invention is not limited to the above-described embodiments, but includes various modifications and equivalent configurations within the spirit and scope of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to configurations including all of the described configurations. Furthermore, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, part of the configuration of each embodiment may be added, deleted, or replaced with other configurations.

[0116] Furthermore, the aforementioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole in hardware, for example by designing them as integrated circuits, or may be realized in software by having a processor interpret and execute a program that realizes each function.

[0117] Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, a hard disk, or an SSD (Solid State Drive), or in a recording medium such as an IC card, an SD card, or a DVD.

[0118] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and do not necessarily show all the control lines and information lines that are necessary for implementation. In reality, it can be considered that almost all components are interconnected. [Explanation of symbols]

[0119] 100...Information processing device 101...Processor 103...Memory 104...Storage section 105...Learning data 106...Correct data 107...Test data 108...Estimated data 109...Learning Model 110...Output section 111...input section 112...Contribution matrix group 113...Pathway data 200...Sample data 203...Drug Data

Claims

1. a computing device that executes predetermined processing; an input unit that receives, as an input, related information relating to each of a plurality of events; a storage unit that stores a learning model that derives an inference result of related information related to the event; an output unit that outputs the estimation result; The computing device calculating an inner product of a first output value of a hidden layer of the learning model to which a first event of the plurality of events is input and a second output value of a hidden layer of the learning model to which a second event of the plurality of events is input; An information processing device characterized by deep learning the learning model based on related information regarding the events so that, for each combination of the events, the difference between the calculated inner product and the correct data for the multiple events is small.

2. 2. The information processing device according to claim 1, the learning model has the same number of layers from the input layer to the hidden layer of the first event as the number of layers from the input layer to the hidden layer of the second event, An information processing apparatus, characterized in that the output value of the hidden layer of the first event and the output value of the hidden layer of the second event have the same number of dimensions.

3. 2. The information processing device according to claim 1, the first event is an analyte; the second event is a drug; The information processing device is characterized in that the predicted result is the therapeutic effectiveness of the drug for the specimen.

4. 4. The information processing device according to claim 3, The hidden layer of the first event represents the degree of genetic abnormality of the specimen; The information processing device is characterized in that the hidden layer of the second event represents the degree of therapeutic effect of a drug.

5. 4. The information processing device according to claim 3, the input unit receives as input specimen-related information relating to a specimen and drug-related information relating to an effect of a drug; The computing device calculating an inner product of a first output value of a hidden layer of the learning model to which the specimen-related information has been input and a second output value of a hidden layer of the learning model to which the drug-related information has been input; An information processing device characterized by deep learning the learning model based on the sample-related information and the drug-related information so that the difference between the calculated inner product and the correct data is small for each combination of the sample and the drug.

6. 2. The information processing device according to claim 1, The information processing device is characterized in that related information relating to the event is expressed using the same feature amount.

7. 6. The information processing device according to claim 5, The computing device calculates a contribution of each of a plurality of items to the inference result using an integrated gradient for the hidden layer; The information processing device is characterized in that the output unit outputs the calculated contribution degree.

8. 8. The information processing device according to claim 7, the calculation device calculates the contribution of the specimen-related information and the drug-related information to the inference result based on an element product of the contribution of each of the calculated plurality of factors; The information processing device is characterized in that the output unit outputs the calculated contribution degree.

9. 8. The information processing device according to claim 7, The association information is a value corresponding to a pathway represented by a graph in which proteins and genes are nodes and the degree of abnormality occurring at each node is an edge, The calculation device calculates a contribution of each pathway to the inference result, The information processing device is characterized in that the output unit outputs the calculated contribution degree together with the pathway.

10. 6. The information processing device according to claim 5, the input unit accepts drug data and specimen data as input; the calculation device calculates the therapeutic effectiveness of the drug data for the specimen data based on the drug data and the specimen data received by the input unit and the learning model; The information processing device is characterized in that the output unit outputs the therapeutic effectiveness.

11. An information processing method executed by an information processing device, The information processing device includes: a computing device that executes predetermined processing; an input unit that receives, as an input, related information relating to each of a plurality of events; a storage unit that stores a learning model that derives an inference result of related information related to the event; an output unit that outputs the estimation result; The information processing method includes: the arithmetic device calculates an inner product of a first output value of a hidden layer of the learning model to which a first event of the plurality of events is input and a second output value of a hidden layer of the learning model to which a second event of the plurality of events is input; An information processing method characterized in that the computing device deep learns the learning model based on related information regarding the events so that, for each combination of the events, the difference between the calculated inner product and the correct data for the multiple events is small.

12. An information processing program for causing an information processing device to execute deep learning of a learning model, The information processing device includes: a computing device that executes predetermined processing; an input unit that receives, as an input, related information relating to each of a plurality of events; a storage unit that stores a learning model that derives an inference result of related information related to the event; an output unit that outputs the estimation result; The information processing program causing the arithmetic device to calculate an inner product of a first output value of a hidden layer of the learning model to which a first event of the plurality of events is input and a second output value of a hidden layer of the learning model to which a second event of the plurality of events is input; An information processing program characterized by causing the computing device to perform deep learning of the learning model based on related information regarding the events so that, for each combination of the events, the difference between the calculated inner product and the correct data for the multiple events is reduced.

Images