Information processing system, and prediction method

Abstract

To enhance prediction accuracy of a therapy-biomarker unknown relation.SOLUTION: An information processing system predicts a therapy-biomarker unknown binary relation from a therapy-biomarker-disease known triadic relation. The information processing system is configured to generate a bipartite graph by diseases expressing the-therapy-biomarker binary relation for each disease from the known triadic relation; calculate, from the bipartite graph by diseases, similarity among therapies by diseases of the therapies, similarity among disease transverse type therapies of the therapies, similarity among biomarkers by diseases of the biomarkers, and similarity among disease transverse type biomarkers of the biomarkers; and calculate and output a prediction score by diseases of an unknown edge, and a disease transverse prediction score.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing system for predicting unknown biomarkers. [Background technology]

[0002] In precision medicine, biomarkers that predict therapeutic effects and side effects play an important role in treatment selection, but it is difficult to predict therapeutic effects and side effects with high accuracy using only the knowledge of known biomarkers. Therefore, there is a need to predict the unknown relationship between treatments and biomarkers.

[0003] The background art of this technical field is a method for constructing a graph based on known relationships and predicting edges that do not exist in the original graph, i.e., unknown relationships.For example, Patent Document 1 (Chinese Patent No. 109033754) describes a method and apparatus for predicting disease-related lncRNAs based on a binary network, which includes the steps of: constructing a binary network based on diseases and lncRNAs according to a dataset of known association relationships between lncRNAs and diseases; calculating disease similarity I and lncRNA similarity I based on common neighborhood; calculating disease similarity II and lncRNA similarity II based on SimRank similarity; obtaining extended disease similarity and extended lncRNA similarity; recycling the extended disease similarity and extended lncRNA similarity into a binary network; and calculating the association degree between diseases and lncRNAs. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Chinese Patent No. 109033754 Summary of the Invention [Problem to be solved by the invention]

[0005] When predicting an unknown relationship between a treatment and a biomarker, since biomarkers are related not only to the treatment but also to the disease, predicting the relationship between the treatment and the biomarker without considering the disease may result in a false relationship being predicted, resulting in reduced prediction accuracy. Therefore, it is necessary to predict the unknown relationship between the treatment and the biomarker with high accuracy by simultaneously considering three types of information: the biomarker, the treatment, and the disease. However, the technology described in Patent Document 1 predicts the unknown relationship between the disease and the lncRNA by generating a bipartite graph having a disease part and an lncRNA part, calculating the similarity between the disease nodes and the lncRNA nodes, and predicting the unknown edge based on the calculated similarity. However, the technology does not consider handling three types of information simultaneously.

[0006] The present invention aims to predict with high accuracy unknown relationships between treatments and biomarkers, taking into account diseases. [Means for solving the problem]

[0007] A representative example of the invention disclosed in the present application is as follows: That is, an information processing system for predicting an unknown binary relationship between a therapy and a biomarker from a known ternary relationship between a therapy, a biomarker, and a disease, the information processing system being configured by a computer having an arithmetic unit for executing predetermined processing and a storage device connected to the arithmetic unit, wherein the arithmetic unit comprises a bipartite graph generator for generating, for each disease, from the known ternary relationship, a disease-specific bipartite graph expressing the binary relationship between the therapy and the biomarker; the disease-specific bipartite graph, the disease-specific treatment similarity, the cross-disease treatment similarity, the disease-specific biomarker similarity, and the cross-disease biomarker similarity; an unknown edge prediction unit in which the arithmetic device calculates at least one of a disease-specific prediction score and a cross-disease prediction score of an unknown edge using the disease-specific bipartite graph, the disease-specific treatment similarity, the cross-disease treatment similarity, the disease-specific biomarker similarity, and the cross-disease biomarker similarity; and an output unit in which the arithmetic device outputs at least one of the disease-specific treatment similarity, the cross-disease treatment similarity, the disease-specific biomarker similarity, the cross-disease biomarker similarity, the disease-specific prediction score, and the cross-disease prediction score. [Effects of the Invention]

[0008] According to one aspect of the present invention, an unknown relationship between a treatment method and a biomarker can be predicted with high accuracy. Problems, configurations, and effects other than those described above will become apparent from the following description of the preferred embodiment of the present invention. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a block diagram showing a configuration of an information processing system according to a first embodiment of the present invention. [Figure 2] FIG. 10 is a diagram showing the configuration of entity set data according to the first embodiment of the present invention. [Figure 3] FIG. 2 is a diagram showing the configuration of ternary relation data according to the first embodiment of the present invention. [Figure 4] 10 is a flowchart of a bipartite graph generation process according to the first embodiment of the present invention. [Figure 5] FIG. 1 is a diagram showing a disease-specific bipartite graph of a disease entity di according to the first embodiment of the present invention. [Figure 6] FIG. 2 is a diagram showing a disease-specific adjacency matrix G(i) according to the first embodiment of the present invention. [Figure 7] 10 is a flowchart of a node similarity calculation process according to the first embodiment of the present invention. [Figure 8] FIG. 1 is a diagram showing a disease-specific treatment-treatment similarity matrix T(i) according to the first embodiment of the present invention. [Figure 9] FIG. 1 is a diagram showing a disease-specific biomarker similarity matrix B(i) according to Example 1 of the present invention. [Figure 10] 10 is a flowchart of an unknown edge prediction process according to the first embodiment of the present invention. [Figure 11] 10 is a flowchart of an output process according to the first embodiment of the present invention. [Figure 12] FIG. 10 is a diagram showing an example of an operation screen that displays an output result in a table format according to the first embodiment of the present invention. [Figure 13] FIG. 10 is a diagram showing an example of an operation screen that displays an output result in a bipartite graph format according to the first embodiment of the present invention. [Figure 14] FIG. 10 is a diagram illustrating the effects of the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0011] Example 1 FIG. 1 is a block diagram showing the configuration of an information processing system according to a first embodiment of the present invention.

[0012] The information processing system of this embodiment includes a server 101 and a database 102. The server 101 and the database 102 are connected so that the server 101 can access the data stored in the database 102.

[0013] The server 101 is a computer having an input device 103, an output device 104, an arithmetic unit 105 for executing a program, a memory 106 for storing the program, and a storage device 107.

[0014] The input device 103 is an interface that receives input to the server 101, such as a mouse, keyboard, or touch panel. The output device 104 is a display device, printer, or the like, and outputs the results of calculations by the calculation device 105 in a format that can be viewed by the user. Note that a terminal (not shown) connected to the server 101 via a network may function as the input device 103 and the output device 104. In this case, the server 101 may have a web server function, and the terminal may access the server 101 using a predetermined protocol (for example, http).

[0015] The arithmetic device 105 is an arithmetic device such as a CPU or a GPU, and executes programs loaded into the memory 106. The arithmetic device 105 executes various programs to realize the various functional units of the server 101 (for example, a bipartite graph generation unit 108, a node-to-node similarity calculation unit 109, an unknown edge prediction unit 110, an output unit 111, etc.). The arithmetic device 105 may include a hardware arithmetic device (for example, an ASIC, an FPGA, etc.).

[0016] The memory 106 includes a ROM, which is a nonvolatile storage element, and a RAM, which is a volatile storage element. The ROM 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 stored in the storage device 107 and data used when the programs are executed.

[0017] The storage device 107 is a non-volatile storage device such as a magnetic storage device (HDD) or a flash memory (SSD), and stores programs executed by the arithmetic device 105 and data used when the programs are executed. Specifically, the storage device 107 stores programs for implementing each of the following units: a bipartite graph generation unit 108, an inter-node similarity calculation unit 109, an unknown edge prediction unit 110, and an output unit 111.

[0018] The bipartite graph generation unit 108 executes a predetermined program to generate a bipartite graph for each disease entity d using the entity set data stored in the entity set data storage unit 112 and the known ternary relation data 300 stored in the known ternary relation data storage unit 113. i A disease-specific adjacency matrix G that represents the edges of a bipartite graph of treatment entities and biomarker entities for (i) Then, we generate a cross-disease adjacency matrix G' that represents the edges of the bipartite graph of treatment entities and biomarker entities ignoring disease information (see Figure 4).

[0019] The node similarity calculation unit 109 executes a predetermined program to calculate the disease-specific adjacency matrix G (i) Using the cross-disease adjacency matrix G', we obtain the disease-specific treatment similarity matrix T (i) and the similarity matrix B between disease-specific biomarkers (i) Then, a cross-disease treatment similarity matrix T' and a cross-disease biomarker similarity matrix B' are calculated (see FIG. 7).

[0020] The unknown edge prediction unit 110 executes a predetermined program to generate a disease-specific adjacency matrix G (i) and the similarity matrix T between disease-specific treatments (i) and the similarity matrix B between disease-specific biomarkers (i) Using the cross-disease treatment similarity matrix T' and the cross-disease biomarker similarity matrix B', the disease-specific prediction adjacency matrix P (i) Then, the cross-disease prediction adjacency matrix P' is calculated (see FIG. 10).

[0021] The output unit 111 executes a predetermined program to visualize and output the calculation results according to the input from the user (see FIG. 11).

[0022] The database 102 includes data to be analyzed by the server 101, that is, an entity set data storage unit 112 (see FIG. 2) and a known triplet relation data storage unit 113 (see FIG. 3).

[0023] The programs executed by the computing device 105 are provided to the server 101 via removable media (CD-ROM, flash memory, etc.) or a network, and are stored in the storage device 107, which is a non-transitory storage medium. For this reason, the server 101 should preferably have an interface for reading data from removable media.

[0024] The server 101 is a computer system configured on one physical computer or on multiple logically or physically configured computers, and may operate on a virtual computer constructed on multiple physical computer resources. For example, each functional unit may operate on a separate physical or logical computer, or multiple functional units may be combined to operate on a single physical or logical computer.

[0025] FIG. 2 is a diagram showing the structure of entity set data stored in the entity set data storage unit 112 according to the first embodiment of the present invention.

[0026] The entity set data includes data for each of a therapy set 201, a biomarker set 202, and a disease set 203.

[0027] The treatment set 201 is a set of L treatment entities {t1, t2, …, t L}. Each of the treatment entities is a drug, surgery, radiation therapy, etc. The treatment entity may be a specific name such as "Ipilimumab" or an abstract name such as "chemotherapy." Note that the definition of the treatment entity may be in a format other than the above example.

[0028] The biomarker set 202 is a set of M biomarker entities {b1, b2, ..., b M Each biomarker entity may be a protein, a gene, an RNA, a clinical test value, TMB (Tumor Mutation Burden), or an intestinal flora. A biomarker entity may be a name such as "ERBB2" or a state such as "overexpression of ERBB2." Note that the definition of a biomarker entity may be in a format other than the above-mentioned examples.

[0029] The disease set 203 is a set of N disease entities {d1, d2, …, d N Each disease entity is the name of a disease (for example, "lung cancer"). Note that the definition of a disease entity may be in a format other than the above example.

[0030] FIG. 3 is a diagram showing the structure of the triplet relation data 300 stored in the known triplet relation data storage unit 113 according to the first embodiment of the present invention.

[0031] The ternary relation data 300 has columns for treatment 301, biomarker 302, and disease 303, and may be represented as tabular data.

[0032] Each row of the ternary relation data 300 represents a ternary relation that is a known association between a therapy entity, a biomarker entity, and a disease entity. For example, the first row 304 represents a known ternary relation where (t1, b1, d1) is a known ternary relation.

[0033] 4 is a flowchart of the bipartite graph generation process according to the first embodiment of the present invention. The bipartite graph generation process shown in FIG.

[0034] Step S401: The bipartite graph generation unit 108 generates a set of treatments {t1, t2, ..., t L}201 and the biomarker set {b1, b2, …, b M}202 and the disease set {d1, d2, …, d N} 203 and known ternary relation data 300 are acquired (S401). The treatment set 201, biomarker set 202, and disease set 203 are acquired from the entity set data storage unit 112, and the known ternary relation data 300 is acquired from the known ternary relation data storage unit 113.

[0035] Step S402: The bipartite graph generation unit 108 generates a bipartite graph for each disease entity d i For example, if the value of the disease column 303 in the known ternary relation data 300 is d i Generate known disease-specific ternary relationship data by extracting rows where

[0036] Step S403: The bipartite graph generation unit 108 generates a bipartite graph for each disease entity d i For the above, disease-specific known binary relation data is generated by removing the disease sequence from the disease-specific known ternary relation data.

[0037] Step S404: The bipartite graph generation unit 108 generates a bipartite graph for each disease entity d i A disease-specific bipartite graph including the treatment section 501 and the biomarker section 502 is generated by using the treatment set 201 as the node 504 of the treatment section 501, the biomarker set 202 as the node 505 of the biomarker section 502, and the binary relationship between the treatment entity and the biomarker entity in the disease-specific known binary relationship data as the edge.

[0038] Step S405: The bipartite graph generation unit 108 generates a bipartite graph for each disease entity d i The disease-specific bipartite graph is expressed as a matrix, and the disease-specific adjacency matrix G (i) Generate.

[0039] Step S406: The bipartite graph generating unit 108 generates known binary relation data by excluding the disease sequence from the known ternary relation data.

[0040] Step S407: The bipartite graph generation unit 108 generates a cross-disease bipartite graph including the therapy section 501 and the biomarker section 502, with the therapy set 201 as the node 504 of the therapy section 501, the biomarker set 202 as the node 505 of the biomarker section 502, and the binary relationship between the therapy entity and the biomarker entity in the known binary relationship data as the edge.

[0041] Step S408: The bipartite graph generation unit 108 generates a cross-disease adjacency matrix G' that represents the cross-disease bipartite graph as a matrix.

[0042] FIG. 5 shows the disease entity d generated in step S404 of the bipartite graph generation process shown in FIG. i FIG. 10 shows a bipartite graph by disease.

[0043] The treatment section 501 represents each treatment entity of the treatment set 201 shown in Fig. 2 as a node 504. The biomarker section 502 represents each biomarker entity of the biomarker set 202 shown in Fig. 2 as a node 505. The treatment-biomarker edge 503 indicates that the treatment entity t1 represented by node 504 and the biomarker entity b1 represented by node 505 are connected to the disease entity d1 generated in step S403 of Fig. 4. i This indicates that the condition is included in the known binary relationship data for each disease.

[0044] The cross-disease bipartite graph generated in step S407 has the same structure as the disease-specific bipartite graph shown in FIG. 5, and only the edges differ between the two.

[0045] Figure 6 shows the disease entity d shown in Figure 5. i A disease-specific adjacency matrix G that represents the edges of the disease-specific bipartite graph for (i) FIG.

[0046] Disease-specific adjacency matrix G (i)is an L-by-M matrix whose rows correspond to therapy entities in the therapy set 201 shown in FIG. 2 and whose columns correspond to biomarker entities in the biomarker set 202 shown in FIG.

[0047] The value of the element in row j and column k of G (i) jk is the treatment entity j and biomarker entity b k but disease entity d i The value G indicates whether the condition is included in the known binary relationship data for each disease. (i) jk If is 1, it indicates a known binary relation, and the value G (i) jk If is 0, it indicates that it is not a known binary relation.

[0048] Disease-specific adjacency matrix G (i) Since the value of row 1, column 1 of is 1, the disease entity d i For , the therapy entity t1 and the biomarker entity b1 are in a known binary relationship.

[0049] The data structure of the cross-disease adjacency matrix G′ generated in step S408 is the same as that of the disease-specific adjacency matrix G shown in FIG. (i) The data structure is the same as that of the previous example, and only the component values ​​differ.

[0050] 7 is a flowchart of the inter-node similarity calculation process according to the first embodiment of the present invention. The inter-node similarity calculation process shown in FIG.

[0051] Step S701: The inter-node similarity calculation unit 109 calculates a disease set {d1, d2, ..., d N}203 and the disease-specific adjacency matrix {G (1) , G (2) , …, G (N)} and the cross-disease adjacency matrix G' is acquired. The disease set 203 is acquired from the entity set data storage unit 112. The disease-specific adjacency matrix and the cross-disease adjacency matrix are generated by the disease-specific bipartite graph generation process.

[0052] Step S702: The node similarity calculation unit 109 calculates the similarity between the nodes of each disease entity d i For the disease-specific adjacency matrix G (i) Using this, the similarity matrix T (i) Calculate the similarity matrix T between disease-specific treatments. (i) is an L-by-L matrix whose rows and columns correspond to the treatment entities of the treatment set 201 shown in FIG. 2, and the value T (i) jk is the disease entity d i About Treatment Entity j and t k This shows the similarity between disease-specific treatments (see Figure 8).

[0053]

number

[0054] Equation 1 is a disease entity d i About the Therapeutic Entity j and t k Similarity T between disease-specific treatments (i) jk This is the calculation formula. G in Equation 1 (i) jk The value of disease entity d i The disease-specific adjacency matrix G (i) The value of the jth row and kth column of f in Equation 1. (i) 1(t j ) is the value of the disease entity d i Therapy entity t in the disease-specific bipartite graph j is the node degree of the node, i.e., the number of edges connected to the node. (i) 2(b k ) is the value of the disease entity d iBiomarker entity b in the disease-specific bipartite graph k In Equation 1, the node degree of the disease entity d generated in step S403 of FIG. i In the disease-specific bipartite graph, the treatment entity t j Node and Treatment Entity t k The greater the number of commonly connected adjacent biomarker nodes between the nodes, the greater the calculated similarity between disease-specific treatments. If there are no commonly connected adjacent biomarker nodes, the similarity between disease-specific treatments will be 0.

[0055] Step S703: The inter-node similarity calculation unit 109 calculates the similarity between the nodes of each disease entity d i For the disease-specific adjacency matrix G (i) Using this, the similarity matrix B between disease-specific biomarkers is (i) Calculate the similarity matrix B between disease-specific biomarkers. (i) is an M-by-M matrix whose rows and columns correspond to the biomarker entities of the biomarker set 202 shown in FIG. 2, and the value B (i) jk is the disease entity d i About Biomarker Entity b j and b k This shows the similarity between disease-specific biomarkers (see Figure 9).

[0056]

number

[0057] Equation 2 is the disease entity d i About biomarker entity b j and b k Similarity B between disease-specific biomarkers (i) jk This is the calculation formula. G in Equation 2 (i) jk , f (i) 1(t j ), f (i) 2(b k) are G in Equation 1, respectively. (i) jk , f (i) 1(t j ), f (i) 2(b k ) in Equation 2. In Equation 2, the disease entity d i In the disease-specific bipartite graph for j Nodes and biomarker entities b k The greater the number of adjacent treatment nodes that are commonly connected to the node, the greater the calculated similarity between disease-specific biomarkers. If there are no commonly connected adjacent treatment nodes, the similarity between disease-specific biomarkers will be 0.

[0058] Step S704: The inter-node similarity calculation unit 109 calculates the disease-specific adjacency matrix {G (1) , G (2) , …, G (N)} and the similarity matrix between disease-specific treatments {T (1) , T (2) ,…, T (N)} to calculate a cross-disease treatment similarity matrix T'. The cross-disease treatment similarity matrix T' is an L-by-L matrix whose rows and columns both correspond to the treatment entities of the treatment set 201 shown in FIG. 2, and the value of the component T' in the jth row and kth column is jk is the treatment entity j and t k The data structure of the cross-disease treatment similarity matrix T' is the disease-specific treatment similarity matrix T shown in Figure 8. (i) The data structure is the same as that of the previous example, but the component values ​​are different.

[0059]

number

[0060] Equation 3 is the treatment entity t j and t k Similarity T' between cross-disease treatments jk In Formula 3, G (i)jk is the disease-specific adjacency matrix G (i) The value of the jth row and kth column of g1(t j ) is the value of the treatment entity t j is the number of disease-specific bipartite graphs with node degrees greater than 0. h1(t j , G (i) ) is calculated by dividing the disease-specific bipartite graph G (i) Treatment Entity j is 1 if the node degree of is greater than 0, otherwise it is 0. j and t k For any disease entity d i Regarding the similarity between disease-specific treatments, T (i) jk If the value of is 0, the cross-disease treatment similarity T' jk The value of will be 0.

[0061] Step S705: The node similarity calculation unit 109 calculates the disease-specific adjacency matrix {G (1) , G (2) , …, G (N)} and the similarity matrix between disease-specific biomarkers {B (1) , B (2) ,…, B (N)} to calculate a cross-disease biomarker similarity matrix B'. The cross-disease biomarker similarity matrix B' is an M-row, M-column matrix whose rows and columns both correspond to the biomarker entities of the biomarker set 202 shown in FIG. 2, and the value B' of the component in the jth row and kth column jk is the biomarker entity b j and b k The data structure of the cross-disease biomarker similarity matrix B' is the same as the disease-specific biomarker similarity matrix B shown in Figure 9. (i) The data structure is the same as that of the previous example, but the component values ​​are different.

[0062]

number

[0063] Equation 4 is a function of the biomarker entity b j and b k Similarity B' between cross-disease biomarkers jk In Equation 4, G (i) jk is the disease-specific adjacency matrix G (i) The value of the jth row and kth column of g2(b k ) is the value of the biomarker entity b k is the number of disease-specific bipartite graphs with node degrees greater than 0. h2(b k , G (i) ) is calculated by dividing the disease-specific bipartite graph G (i) Biomarker entity b k is 1 if the node degree of is greater than 0, otherwise it is 0. j and b k For any disease entity d i Similarity between disease-specific biomarkers B (i) jk If the value of is 0, the cross-disease biomarker similarity B' jk The value of will be 0.

[0064] FIG. 8 shows the disease-specific treatment similarity matrix T (i) FIG.

[0065] Similarity matrix T between disease-specific treatments (i) is an L-by-L matrix whose rows and columns correspond to the treatment entities of the treatment set 201 shown in FIG. 2, and the value T (i) jk is the disease entity d i About Treatment Entity j and t k This represents the similarity between disease-specific treatments.

[0066] Similarity matrix T between disease-specific treatments (i) The component 801 in the first row and first column of iindicates that the similarity between disease-specific treatments between treatment entities t1 and t1 is 0.8.

[0067] FIG. 9 shows the disease-specific biomarker similarity matrix B generated in step S703 of the node similarity calculation process shown in FIG. (i) FIG.

[0068] Similarity matrix B between disease-specific biomarkers (i) is an M-by-M matrix whose rows and columns correspond to the biomarker entities of the biomarker set 202 shown in FIG. 2, and the value B (i) jk is the disease entity d i About Biomarker Entity b j and b k This represents the degree of similarity between disease-specific biomarkers.

[0069] Similarity matrix B between disease-specific biomarkers (i) The component 901 in the first row and first column of i represents that the disease-specific biomarker similarity between biomarker entities b1 and b1 is 0.5.

[0070] FIG. 10 is a flowchart of the unknown edge prediction process according to the first embodiment of the present invention.

[0071] This unknown edge prediction process is executed by the unknown edge prediction unit 110 of the server 101.

[0072] Step S1001: The unknown edge prediction unit 110 predicts the disease set {d1, d2, ..., d N}203 and the disease-specific adjacency matrix {G (1) , G (2) , …, G (N)} and the similarity matrix between disease-specific treatments {T (1) , T (2) ,…, T (N)} and the disease-specific biomarker similarity matrix {B (1) , B (2) ,…, B(N)}, a cross-disease treatment similarity matrix T', and a cross-disease biomarker similarity matrix B' are acquired. The disease set 203 is acquired from the entity set data storage unit 112. The disease-specific adjacency matrix is ​​generated in the disease-specific bipartite graph generation process. The disease-specific treatment similarity matrix, disease-specific biomarker similarity matrix, cross-disease treatment similarity matrix, and cross-disease biomarker similarity matrix are generated by the node-to-node similarity calculation process.

[0073] Step S1002: The unknown edge prediction unit 110 calculates the unknown edge of each disease entity d i For the disease-specific adjacency matrix G (i) and the similarity matrix T between disease-specific treatments (i) and the similarity matrix T' between cross-disease treatments and the similarity matrix B between disease-specific biomarkers. (i) Using the cross-disease biomarker similarity matrix B', we obtain the disease-specific prediction adjacency matrix P (i) Calculate the disease-specific predicted adjacency matrix P (i) is an L-by-M matrix whose rows correspond to the treatment entities of the treatment set 201 shown in FIG. 2 and whose columns correspond to the biomarker entities of the biomarker set 202 shown in FIG. 2. The disease-specific prediction adjacency matrix P (i) The value of the jth row and kth column of P (i) jk is the disease entity d i About Treatment Entity j and biomarker entity b k The disease-specific prediction score P (i) jk The larger the value of , the more likely the disease entity d i About Therapeutic Entity j and biomarker entity b k It indicates that there is a high possibility that a binary relationship of the form P exists. Note that the disease-specific prediction score is calculated for any binary relationship without distinguishing between known and unknown binary relationships. The disease-specific prediction adjacency matrix P (i) The data structure of the disease-specific adjacency matrix G shown in Figure 6 (i)The data structure is the same as that of the previous example, but the component values ​​are different.

[0074]

number

[0075] Equation 5 is the disease-specific prediction adjacency matrix P (i) In Equation 5, p, q, u, and v are hyperparameters that take any value between 0 and 1, and determine the contribution rate of each matrix. Specifically, p is the disease-specific treatment similarity matrix T (i) q is the cross-disease treatment similarity weight that adjusts the contribution rate of the cross-disease treatment similarity matrix T'. u is the disease-specific biomarker similarity matrix B (i) v is the disease-specific biomarker similarity weight that adjusts the contribution rate of the cross-disease biomarker similarity matrix B'.

[0076] When the sum of the similarity weight p between disease-specific treatments and the similarity weight u between disease-specific biomarkers is set to 1, the similarity matrix T (i) and the similarity matrix B between disease-specific biomarkers (i) Only the disease-specific prediction adjacency matrix P (i) In this case, the disease entity d i Any disease entity d different from l Similarity information about disease entities d l The information on the edges of the disease-specific bipartite graph is used as the disease-specific predicted adjacency matrix P (i) Since the calculated value of does not affect the false edge, the disease-specific prediction score is small, which means that the prediction specificity is high. i Any disease entity d different from l Since the edge information of the disease-specific bipartite graph for is not used, the disease-specific prediction score is 0, meaning that there are many edges that are essentially impossible to predict, and the prediction sensitivity is low.

[0077] When the sum of the cross-disease treatment similarity weight q and the cross-disease biomarker similarity weight v is set to 1, only the cross-disease treatment similarity matrix T' and the cross-disease biomarker similarity matrix B' are included in the disease-specific prediction adjacency matrix P (i) In this case, the disease entity d i Any disease entity d that is different from l Similarity information about disease entities d l The information on the edges of the disease-specific bipartite graph is used as the disease-specific predicted adjacency matrix P (i) This affects the calculated value of , so the number of edges with disease-specific prediction scores greater than 0 increases, meaning that the sensitivity of the prediction is high. In addition, the number of false edges also increases, meaning that the specificity of the prediction is low.

[0078] In other words, the sensitivity and specificity of the prediction can be adjusted by adjusting the hyperparameter values.

[0079] Step S1003: The unknown edge prediction unit 110 calculates the disease-specific predicted adjacency matrix {P (1) , P (2) ,…, P (N)} is used to calculate the cross-disease predictive adjacency matrix P'. The cross-disease predictive adjacency matrix P' is an L-row, M-column matrix whose rows correspond to therapy entities in the therapy set 201 shown in FIG. 2 and whose columns correspond to biomarker entities in the biomarker set 202 shown in FIG. 2. The value P' of the element in the jth row and kth column of the cross-disease predictive adjacency matrix P' is jk is the treatment entity j and biomarker entity b k The cross-disease prediction score P' is a binary relationship between the two. jk The larger the value of , the more likely it is that for some disease entity, the more likely it is that the treatment entity t j and biomarker entity b k The cross-disease prediction score is calculated for any binary relationship without distinguishing between known and unknown binary relationships. The data structure of the cross-disease prediction adjacency matrix P' is the same as that of the disease-specific adjacency matrix G shown in Figure 6.(i) The data structure is the same as that of the previous example, but the component values ​​are different.

[0080]

number

[0081] Equation 6 is a treatment entity t j and biomarker entity b k Dyadic cross-disease prediction score P' jk In Equation 6, {P (1) jk , P (2) jk , …, P (N) jk} maximum value of P' jk The value is set to

[0082] FIG. 11 is a flowchart of the output process according to the first embodiment of the present invention.

[0083] This output process is mainly executed by the output unit 111 of the server 101.

[0084] Step S1101: The user inputs values ​​of the disease-specific treatment similarity weight p, the cross-disease treatment similarity weight q, the disease-specific biomarker similarity weight u, and the cross-disease biomarker similarity weight v into the input device 103.

[0085] Step S1102: Based on the values ​​entered by the user in step S1101, a disease-specific predicted adjacency matrix {P (1) , P (2) ,…, P (N)} and calculates the cross-disease predicted adjacency matrix P′. The unknown edge prediction process is executed by the unknown edge prediction unit 110 of the server 101.

[0086] Step S1103: The user inputs the display format of the results in the display format input area 1206 (see FIGS. 12 and 13). The input value is "table format" or "bipartite graph format." If the input value is "table format," the process proceeds to step S1104, and the operation screen shown in FIG. 12 is presented to the user. If the input value is "bipartite graph format," the process proceeds to step S1107, and the operation screen shown in FIG. 13 is presented to the user.

[0087] Step S1104: The user inputs a treatment entity or a biomarker entity for which results are to be displayed in the entity input area 1213 (see FIG. 12). The input value is a treatment set {t1, t2, ..., t L}201 or any element of the biomarker set 202 {b1, b2, …, b M} element.

[0088] Step S1105: Based on the value entered by the user in step S1104, data to be displayed is extracted.

[0089] For example, if the input value is a treatment entity t j If so, the j-th row of the cross-disease treatment similarity matrix T' and each disease entity d i Similarity matrix T between disease-specific treatments (i) The j-th row of the cross-disease prediction adjacency matrix P' and each disease entity d i Disease-specific prediction adjacency matrix P (i) The j-th row of the cross-disease adjacency matrix G' and each disease entity d i The disease-specific adjacency matrix G (i) Extract the jth row of and.

[0090] Also, for example, if the input value is a biomarker entity b k If the k-th column of the cross-disease biomarker similarity matrix B' and each disease entity d i Similarity matrix B between disease-specific biomarkers (i) The k-th column of the cross-disease prediction adjacency matrix P' and each disease entity d iDisease-specific prediction adjacency matrix P (i) The k-th column of the cross-disease adjacency matrix G' and each disease entity d i The disease-specific adjacency matrix G (i) Extract the kth column of and.

[0091] Step S1106: The output unit 111 displays the data extracted in step S1105 in a table format (see FIG. 12).

[0092] Step S1107: If the input value is in the "bipartite graph format", the user inputs the treatment entity or biomarker entity for which the results are to be displayed in the entity input area 1213 (see FIG. 13). The input value is a treatment set {t1, t2, ..., t L}201 or any element of the biomarker set 202 {b1, b2, …, b M} element.

[0093] Step S1108: The user inputs the category of the bipartite graph. The category of the bipartite graph is "disease-specific" or "cross-disease." If the input value is "disease-specific," the process proceeds to step S1109, where the disease-specific bipartite graph is displayed. If the input value is "cross-disease," the process proceeds to step S1110, where the cross-disease bipartite graph is displayed.

[0094] Step S1109: The user inputs a disease entity for which the results are to be displayed. The input value is a disease set {d1, d2, ..., d N}203. The input value is d i If so, the displayed bipartite graph is a disease entity d i This is a bipartite graph by disease.

[0095] Step S1110: The user inputs a similarity display threshold and a predicted score display threshold.

[0096] Step S1111: The output unit 111 extracts data for displaying the results in a bipartite graph format based on the values ​​input by the user from step S1107 to step S1110.

[0097] For example, if the input value in step S1107 is t j , the input value in step S1108 is "disease type", and the input value in step S1109 is d i The case where the input value t j The cross-disease treatment similarity between is equal to or greater than the similarity display threshold, or the disease entity d i For input value t j All treatment entities whose disease-specific treatment similarity between them is equal to or greater than the similarity display threshold are extracted as treatment nodes to be displayed in the bipartite graph. At this time, the treatment similarity is extracted to reflect the thickness of the border of the treatment node. Next, the disease entity d i For input value t j All biomarker entities whose disease-specific prediction scores between them are equal to or greater than the prediction score display threshold are extracted as biomarker nodes to be displayed in the bipartite graph. At this time, the prediction scores are extracted to reflect the thickness of the edges of the bipartite graph. Next, whether or not a known binary relationship exists between the extracted treatment entities and biomarker entities is determined based on the disease entity d in order to reflect the line type of the edges of the bipartite graph and the line type of the biomarker nodes. i The disease-specific adjacency matrix G (i) For example, edges corresponding to known binary relations can be displayed as solid lines, and edges corresponding to unknown binary relations can be displayed as dashed lines. j The borders of biomarker nodes with which a known binary relationship exists are displayed with solid lines, and the borders of biomarker nodes with which no known binary relationship exists are displayed with dashed lines.

[0098] For example, if the input value in step S1107 is t j The case where the input value in step S1108 is “cross-disease type” will be described. jAll treatment entities whose cross-disease treatment similarity between t and t is equal to or greater than the similarity display threshold are extracted as treatment nodes to be displayed in the bipartite graph. In addition, the treatment similarity is extracted to reflect the thickness of the border of the treatment node. Next, the input value t j All biomarker entities whose cross-disease prediction scores between them are equal to or greater than the prediction score display threshold are extracted as biomarker nodes to be displayed in the bipartite graph. In addition, the prediction scores are extracted to reflect the thickness of the edges of the bipartite graph. Next, whether or not there is a known binary relationship between the extracted treatment entity and biomarker entity is extracted from the cross-disease adjacency matrix G' to reflect this in the line type of the edges of the bipartite graph and the line type of the biomarker node. For example, edges corresponding to known binary relationships may be displayed with solid lines, and edges corresponding to unknown binary relationships may be displayed with dashed lines. The input value t j The borders of biomarker nodes with which a known binary relationship exists are displayed with solid lines, and the borders of biomarker nodes with which no known binary relationship exists are displayed with dashed lines.

[0099] Also, for example, if the input value in step S1107 is b k The input value in step S1108 is "disease-specific" and the input value in step S1109 is d i The case where the input value b k The cross-disease biomarker similarity between is equal to or greater than the similarity display threshold, or the disease entity d i Regarding input value b k All biomarker entities whose disease-specific biomarker similarity between is equal to or greater than the similarity display threshold are extracted as biomarker nodes to be displayed in the bipartite graph. In addition, the biomarker similarity is extracted to reflect the thickness of the border of the biomarker node. Next, the input value b k Disease entities between iAll treatment entities whose disease-specific prediction scores are equal to or greater than the prediction score display threshold are extracted as treatment nodes to be displayed in the bipartite graph. The prediction scores are also extracted to reflect the thickness of the edges in the bipartite graph. Next, whether or not a known binary relationship exists between the extracted treatment entities and biomarker entities is determined based on the disease entity d to reflect the line type of the edges in the bipartite graph and the line type of the biomarker node. i The disease-specific adjacency matrix G (i) For example, edges corresponding to known binary relations can be displayed as solid lines, and edges corresponding to unknown binary relations can be displayed as dashed lines. k The borders of treatment nodes with which a known binary relationship exists may be displayed with solid lines, and the borders of treatment nodes with which no known binary relationship exists may be displayed with dashed lines.

[0100] Also, for example, if the input value in step S1107 is b k The case where the input value b in step S1108 is "cross-disease type" will be described. k All biomarker entities whose cross-disease biomarker similarity between is equal to or greater than the similarity display threshold are extracted as biomarker nodes to be displayed in the bipartite graph. In addition, the biomarker similarity is extracted to reflect the thickness of the border of the biomarker node. Next, the input value b k All treatment entities whose cross-disease prediction scores between them are equal to or greater than the prediction score display threshold are extracted as treatment nodes to be displayed in the bipartite graph. In addition, the prediction scores are extracted to reflect the thickness of the edges in the bipartite graph. Next, whether or not there is a known binary relationship between the extracted treatment entities and biomarker entities is extracted from the cross-disease adjacency matrix G' to reflect this in the line type of the edges in the bipartite graph and the line type of the biomarker nodes. For example, edges corresponding to known binary relationships may be displayed with solid lines, and edges corresponding to unknown binary relationships may be displayed with dashed lines. Input value b k The borders of treatment nodes with which a known binary relationship exists may be displayed with solid lines, and the borders of treatment nodes with which no known binary relationship exists may be displayed with dashed lines.

[0101] Step S1112: The output unit 111 displays the data extracted in step S1111 in the form of a bipartite graph (see FIG. 13).

[0102] FIG. 12 is a diagram showing an example of an operation screen that displays the results output in the output process according to the first embodiment of the present invention in a table format.

[0103] The user can comprehensively refer to the calculation results through the operation screen shown in Figure 12. For example, it can be used by pharmaceutical researchers to search for unknown biomarkers for a treatment they are interested in. However, the method of use is not limited to this example.

[0104] The operation screen shown in FIG. 12 includes an unknown edge prediction execution area 1201, a display format selection area 1202, and a result display area 1203.

[0105] The unknown edge prediction execution area 1201 includes a weight input area 1204 and an unknown edge prediction execution button 1205 .

[0106] First, the user inputs the value of the disease-specific treatment similarity weight p, the value of the cross-disease treatment similarity weight q, the value of the disease-specific biomarker similarity weight u, and the value of the cross-disease biomarker similarity weight v into the weight input area 1204. Only values ​​equal to or greater than 0 can be input as weight values.

[0107] Next, when the user operates the unknown edge prediction execution button 1205, the unknown edge prediction unit 110 of the server 101 executes the unknown edge prediction process based on the values ​​of the weights p, q, u, and v entered in the weight input area 1204. At this time, the values ​​of the weights p, q, u, and v entered in the weight input area 1204 are normalized to p / (p+q+u+v), q / (p+q+u+v), u / (p+q+u+v), and v / (p+q+u+v), respectively, so that the sum p+q+u+v becomes 1. Note that if the values ​​of the weights p, q, u, and v entered in the weight input area 1204 are all 0, the unknown edge prediction process is not executed, and a prompt may be displayed to prompt the user to make one or more of the weight values ​​p, q, u, and v positive.

[0108] The display format selection area 1202 includes a display format input area 1206 and a display format decision button 1207. The display format selection area 1202 is displayed after the unknown edge prediction execution button 1205 is operated and the unknown edge prediction process is executed.

[0109] First, the user inputs a display format in the display format input area 1206. The input display format value is "table" or "bipartite graph." When "table" is input in the display format input area 1206, the results are displayed in the result display area 1203 in table format. When "bipartite graph" is input in the display format input area 1206, the results are displayed in the result display area 1203 in bipartite graph format (see FIG. 13). Next, by operating the display format determination button 1207, the results are displayed in the result display area 1203. FIG. 12 illustrates an example in which "table" is input in the display format input area 1206.

[0110] 12 displays the results in a table format. When the results are displayed in a table format, the result display area 1203 includes an entity selection area 1208, a table display button 1209, a similarity display area 1210, and a predicted score display area 1211. The similarity display area 1210 and the predicted score display area 1211 are displayed after the table display button 1209 is operated.

[0111] The entity selection area 1208 includes an entity category input area 1212 and an entity input area 1213 .

[0112] First, the user inputs the entity category in the entity category input area 1212. The entity category value is either "therapy" or "biomarker." Next, the user inputs the entity in the entity input area 1213 and inputs the specific therapy or biomarker to be displayed. If the entity category is "therapy," the entity is a set of therapies {t1, t2, ..., t L} 201. If the category of the entity is "biomarker", the entity is a member of the biomarker set 202 {b1, b2, ..., b M}. In Figure 12, the entity category is "treatment" and the entity is j This shows an example in which the following is input.

[0113] Next, the user operates the table display button 1209. As a result, results related to the entity entered in the entity selection area 1208 are displayed in a similarity display area 1210 and a predicted score display area 1211.

[0114] The similarity display area 1210 displays the set of treatments {t1, t2, ..., t L}201 Treatment Entity t k and the disease set {d1, d2, …, d M}203 disease entities di About Therapeutic Entity j and t k The cross-disease treatment similarity 1214 between the disease entities d i About Treatment Entity j and t k If "biomarker" is entered in the entity category input area 1212, the similarity display area 1210 displays the similarity 1215 between the disease-specific treatments. M Each of the 202 biomarker entities b k and the disease set {d1, d2, …, d M}203 disease entities d i For biomarker entity b j and b k The cross-disease biomarker similarity between the disease entities d i About Biomarker Entity b j and b k The similarity between disease-specific biomarkers between the above two disease-specific biomarkers is displayed. The data in the similarity display area 1210 can be sorted according to the values ​​of any one or more columns.

[0115] Treatment Entity j and t k The value of the cross-disease treatment similarity 1214 between the two is the component value T' of the jth row and kth column of the cross-disease treatment similarity matrix T'. jk is.

[0116] Disease Entity d i About Treatment Entity j and t k The value of the similarity 1215 between disease-specific treatments is the similarity matrix T (i) The component value T in the jth row and kth column of (i) jk is.

[0117] The predicted score display area 1211 displays the biomarker set {b1, b2, ..., b M Each of the 202 biomarker entities bk About Therapeutic Entity j and biomarker entity b k The cross-disease prediction score 1216 between the disease set {d1, d2, …, d L}203 for all disease entities, treatment entities t j and biomarker entity b k Whether the binary relation between d1, d2, …, d L}203 disease entities d i About Treatment Entity j and biomarker entity b k The disease-specific prediction score 1218 between the disease set {d1, d2, …, d L}203 disease entities d i About Therapeutics Entity j and biomarker entity b k If "biomarker" is entered in the entity category input area 1212, the predicted score display area 1211 displays the set of treatments {t1, t2, ..., t L}201 Treatment Entity t j For biomarker entity b k and treatment entities j and the cross-disease prediction score between the disease set {d1, d2, …, d L} Biomarker entity b for all 203 disease entities k and treatment entities j whether the binary relation between the disease set {d1, d2, …, d L}203 disease entities d i About Biomarker Entity b k and treatment entities j and the disease-specific prediction scores between the disease set {d1, d2, …, d L}203 disease entities d i About Biomarker Entity b kand treatment entities j It is displayed whether the binary relation between the two is unknown. The data in the predicted score display area 1211 can be sorted according to the values ​​of any one or more columns.

[0118] Treatment Entity j and biomarker entity b k The value of the cross-disease prediction score 1216 between jk is.

[0119] All disease entities d in the disease set 203 i Treatment Entity j and biomarker entity b k The value of 1217 is the value G' in the jth row and kth column of the cross-disease adjacency matrix. jk If it is 0, the answer is "yes"; otherwise, the answer is "no."

[0120] Disease Entity d i Treatment Entity j and biomarker entity b k The values ​​of the disease-specific prediction scores 1218 between (i) The component value P in the jth row and kth column of (i) jk is.

[0121] Disease Entity d i Treatment Entity j and biomarker entity b k The value of 1219 is unknown whether the binary relation of disease entity d i The value of jth row and kth column of the disease-specific adjacency matrix G (i) jk If it is 0, the answer is "yes"; otherwise, the answer is "no."

[0122] 12 displays all of the disease-specific treatment similarity, cross-disease treatment similarity, disease-specific biomarker similarity, cross-disease biomarker similarity, disease-specific prediction score, and cross-disease prediction score, but any one of them may be displayed depending on the application. Also, while the example of the operation screen shown in FIG. 12 displays both treatments and biomarkers, either treatments or biomarkers may be displayed depending on the application.

[0123] FIG. 13 is a diagram showing an example of an operation screen that displays the results output in the output process according to the first embodiment of the present invention in the form of a bipartite graph.

[0124] Through the operation screen shown in FIG. 13, the user can easily visually recognize known and unknown binary relations in the form of a bipartite graph, and can also adjust the nodes and edges displayed in the bipartite graph by threshold processing.

[0125] The operation screen shown in Fig. 13 includes an unknown edge prediction execution area 1201, a display format selection area 1202, and a result display area 1203. Fig. 13 illustrates an example in which "bipartite graph" is input in the display format input area 1206, and the result is displayed in the bipartite graph format in the result display area 1203.

[0126] The unknown edge prediction execution area 1201 and the display format selection area 1202 are the same as those in FIG. 12, and therefore their explanation will be omitted.

[0127] 13 displays results in bipartite graph format. When results in bipartite graph format are displayed, result display area 1203 includes entity selection area 1208, bipartite graph selection area 1301, threshold setting area 1302, bipartite graph display button 1303, and bipartite graph display area 1304. Bipartite graph display area 1304 is displayed after operating bipartite graph display button 1303.

[0128] The entity selection area 1208 is the same as that in FIG. 12, and therefore the explanation will be omitted. In FIG. 13, the entity category is set to "treatment" and the entity is set to "t". j is input.

[0129] The bipartite graph selection area 1301 includes a bipartite graph category input area 1305 and a disease entity input area 1306 .

[0130] The user inputs the category of the bipartite graph in the bipartite graph category input area 1305. The value of the bipartite graph category is "disease-specific" or "cross-disease". If the category of the bipartite graph is "disease-specific", then the user inputs a disease entity in the disease entity input area 1306. The disease entity is a disease set {d1, d2, ..., d N}203. In FIG. 13, the category of the bipartite graph is "Disease" and the disease entity is d i is input.

[0131] The threshold setting area 1302 includes a similarity display threshold input area 1307 and a predicted score display threshold input area 1308. Fig. 13 illustrates an example in which 0.2 is input in the similarity display threshold input area 1307 and 0.3 is input in the predicted score display threshold input area 1308.

[0132] After inputting information into the entity selection area 1208 , bipartite graph selection area 1301 , and threshold setting area 1302 , the user can display the bipartite graph in the bipartite graph display area 1304 by operating the bipartite graph display button 1303 .

[0133] In the bipartite graph display area 1304, the treatment entity t in the entity selection area 1208 is displayed. j In the bipartite graph selection area 1301, select "Disease" as the bipartite graph category and select d as the disease entity. i Since the disease entity d iAbout Treatment Entity j If "cross-disease" is entered as the category of the bipartite graph, the bipartite graph display area 1304 displays the treatment entity t j A cross-disease bipartite graph for

[0134] Therapy section 1309 includes a therapy entity t j Cross-disease treatment similarity 1311 between, or disease entity d i About Treatment Entity j The nodes of the treatment entities are displayed if the similarity 1312 between the treatments by disease and the treatment entity t is equal to or greater than 0.2, which is the value entered in the similarity display threshold input area 1307. j The cross-disease treatment similarity 1311 between the disease entity d i About Treatment Entity j A disease-specific treatment similarity 1312 between the two is displayed. The border of each treatment node may be drawn thicker as the average values ​​of the cross-disease treatment similarity 1311 and the disease-specific treatment similarity 1312 increase.

[0135] The biomarker section 1310 includes disease entities d i Treatment Entity j The nodes of biomarker entities are displayed if the disease-specific prediction score 1313 between the treatment entity t and the treatment entity t is equal to or greater than 0.3, which is the value entered in the prediction score display threshold input area 1308. j The disease-specific prediction score 1313 between the treatment entity t and the biomarker node is displayed. j If a known binary relationship exists, it should be displayed as a solid line, and if no known binary relationship exists, it should be displayed as a dashed line.

[0136] An edge 1314 in the bipartite graph connects the disease entity d between the treatment entity and the biomarker entity that the edge connects. iThe larger the disease-specific prediction score, the thicker the line should be drawn. In addition, if there is a known binary relationship between the treatment entity and the biomarker entity connected by the edge, it should be displayed as a solid line, and if there is no known binary relationship, it should be displayed as a dashed line.

[0137] 13, all of the disease-specific treatment similarity, cross-disease treatment similarity, disease-specific biomarker similarity, cross-disease biomarker similarity, disease-specific prediction score, and cross-disease prediction score are displayed, but any one of them may be displayed depending on the application. Also, the example of the operation screen shown in FIG. 13 displays both treatments and biomarkers, but either treatments or biomarkers may be displayed depending on the application.

[0138] As described above, the information processing system of Example 1 generates a disease-specific bipartite graph of treatments and biomarkers, calculates disease-specific node-to-node similarities and cross-disease node-to-node similarities from the generated disease-specific bipartite graph, and performs edge prediction using these node-to-node similarities in combination, thereby making it possible to predict unknown relationships between treatments and biomarkers with high accuracy.

[0139] Furthermore, predicting a binary relationship between a disease-specific treatment and a biomarker is essentially equivalent to predicting a ternary relationship between a treatment, a biomarker, and a disease. In other words, by using a cross-disease predictive adjacency matrix, it is possible to predict a binary relationship between a treatment and a biomarker, and by using a disease-specific predictive adjacency matrix, it is possible to predict a ternary relationship between a treatment, a biomarker, and a disease. Note that, although the prediction of an unknown relationship between a treatment and a biomarker has been described in Example 1, the information processing system of Example 1 can be used to support the selection of a treatment suitable for the patient's symptoms based on the unknown relationship predicted by Example 1 after receiving biomarker information or treatment information of a patient as input during a doctor's examination.

[0140] <Example 2> In this embodiment, the specificity of edge prediction is reduced and the sensitivity of edge prediction is increased. In this embodiment, only the configuration and processing different from the above-mentioned embodiment 1 will be described with reference to Figures 1 to 13, and the description of the configuration and processing that are the same as those in embodiment 1 will be omitted.

[0141] In the inter-node similarity calculation process shown in FIG. 7, the method of calculating the cross-disease treatment similarity matrix in step S704 and the method of calculating the cross-disease biomarker similarity matrix in step S705 are different from those in the first embodiment.

[0142] Step S704: The inter-node similarity calculation unit 109 calculates a cross-disease treatment similarity matrix T' using the cross-disease adjacency matrix G'. The cross-disease treatment similarity matrix T' is an L-row, L-column matrix whose rows and columns both correspond to the treatment entities of the treatment set 201 shown in Figure 2, and the value T' of the j-th row and k-th column component jk is the treatment entity j and t k The data structure of the cross-disease treatment similarity matrix T' is the disease-specific treatment similarity matrix T shown in Figure 8. (i) The data structure is the same as that of the previous example, but the component values ​​are different.

[0143]

number

[0144] Equation 7 is a treatment entity t j and t k Similarity T' between cross-disease treatments jk This is the calculation formula for G' in Equation 7. jk The value of f'1(t j ) is the value of the treatment entity t in the transdisease bipartite graph. j is the node degree of f'2(b k ) is the value of the biomarker entity b in the cross-disease bipartite graph. kIn Equation 7, in the transdisease bipartite graph generated in step S407 of FIG. 4, the treatment entity t j Node and Treatment Entity t k The greater the number of commonly connected adjacent biomarker nodes, the greater the calculated similarity. If there are no commonly connected adjacent biomarker nodes, the cross-disease treatment similarity will be 0.

[0145] Step S705: The inter-node similarity calculation unit 109 calculates a cross-disease biomarker similarity matrix B' using the cross-disease adjacency matrix G'. The cross-disease biomarker similarity matrix B' is an M-row, M-column matrix whose rows and columns both correspond to the biomarker entities of the biomarker set 202 shown in FIG. 2, and the value B' of the j-th row and k-th column component jk is the biomarker entity b j and b k The data structure of the cross-disease biomarker similarity matrix B' is the same as the disease-specific biomarker similarity matrix B shown in Figure 9. (i) The data structure is the same as that of the previous example, but the component values ​​are different.

[0146]

number

[0147] Equation 8 is a function of the biomarker entity b j and b k Similarity B' between cross-disease biomarkers jk This is the calculation formula for G' in Equation 8. jk , f'1(t j ), f'2(b k ) are G' in Equation 7, respectively. jk , f'1(t j ), f'2(b k ) in the cross-disease bipartite graph generated in step S407 of FIG. 4. j Nodes and biomarker entities b kThe greater the number of adjacent treatment nodes that are commonly connected to a node, the greater the calculated similarity. If there are no commonly connected adjacent treatment nodes, the cross-disease biomarker similarity will be 0.

[0148] In step S704, in the first embodiment, the disease-specific adjacency matrix {G (1) , G (2) , …, G (N)} and the similarity matrix between disease-specific treatments {T (1) , T (2) ,…, T (N)}, a cross-disease treatment similarity matrix T′ is calculated using Equation 3. In Example 1, a treatment entity t j and t k For any disease entity d i About T (i) jk If the value of is 0, T' jk On the other hand, in Example 2, the cross-disease adjacency matrix G' is used to calculate the cross-disease treatment similarity matrix T' using Equation 7. In Example 2, the cross-disease adjacency matrix generated ignoring disease information is used, so that the value of j and t k For any disease entity d i About T (i) jk Even if the value of is 0, T' jk The value of may be greater than 0. Therefore, the number of non-zero elements in the cross-disease treatment similarity matrix T' calculated in Example 2 is equal to or greater than the number of non-zero elements in the cross-disease treatment similarity matrix T' calculated in Example 1. Disease-specific predicted adjacency matrix P (i) is calculated using Equation 5, the more non-zero elements there are in the cross-disease treatment similarity matrix T', the more the disease-specific predicted adjacency matrix P (i) It is expected that the number of non-zero elements of , is large. This may lead to a decrease in the specificity of edge prediction, but may also lead to an increase in sensitivity.

[0149] In step S705, in the first embodiment, the disease-specific adjacency matrix {G (1), G (2) , …, G (N)} and the disease-specific biomarker similarity matrix {B (1) , B (2) ,…, B (N)}, a cross-disease biomarker similarity matrix B′ is calculated using Equation 4. In Example 1, a certain biomarker entity b j and b k For any disease entity d i About B (i) jk If the value of is 0, then B' jk The value of is 0. On the other hand, in Example 2, the cross-disease adjacency matrix G' is calculated using Equation 8 based on the cross-disease adjacency matrix G'. In Example 2, since the cross-disease adjacency matrix generated ignoring disease information is used, the value of a certain biomarker entity b j and b k For any disease entity d i About B (i) jk Even if the value of B' is 0, jk The value of may be greater than 0. Therefore, the number of non-zero elements in the cross-disease biomarker similarity matrix B' calculated in Example 2 is equal to or greater than the number of non-zero elements in the cross-disease biomarker similarity matrix B' calculated in Example 1. Disease-specific predicted adjacency matrix P (i) is calculated using Equation 5, the more non-zero elements there are in the cross-disease biomarker similarity matrix B', the greater the disease-specific predicted adjacency matrix P (i) It is expected that the number of non-zero elements of , is large. This may lead to a decrease in the specificity of edge prediction, but may also lead to an increase in sensitivity.

[0150] As described above, in the information processing system of Example 2, by changing the method for calculating the cross-disease treatment similarity matrix in step S704 in Example 1 and the method for calculating the cross-disease biomarker similarity matrix in step S705 in Example 1, it is possible to reduce the specificity of the relationship prediction and increase the sensitivity of the relationship prediction.

[0151] The unknown binary relationship between a treatment and a biomarker predicted by the information processing systems of Examples 1 and 2 described above can be used in clinical practice to support the selection of a treatment based on a biomarker detected in a patient. Furthermore, by inputting the unknown binary relationship between a predicted treatment and a biomarker into a treatment effect prediction system separate from the present invention, it can be used to predict the degree of treatment effect.

[0152] Next, the effects of the information processing systems of Examples 1 and 2 will be explained in comparison with a conventional information processing system that does not take disease information into account. In the following description, the information processing system of Example 1 will be referred to as "Proposed Method 1," the information processing system of Example 2 will be referred to as "Proposed Method 2," and the conventional information processing system that does not take disease information into account will be referred to as "Conventional Method."

[0153] In the conventional method, the cross-disease adjacency matrix G', the cross-disease treatment similarity matrix T', and the cross-disease biomarker similarity matrix B' are calculated using the same procedure as in proposed method 2. On the other hand, in the conventional method, the disease-specific adjacency matrix {G (1) , G (2) , …, G (N)}, disease-specific treatment method similarity matrix {T (1) , T (2) ,…, T (N)}, similarity matrix between disease-specific biomarkers {B (1) , B (2) ,…, B (N)}, and disease-specific prediction adjacency matrix {P (1) , P (2) ,…, P (N)} is not calculated. Furthermore, the method for calculating the cross-disease predicted adjacency matrix P' in the conventional method differs from the method for calculating the cross-disease predicted adjacency matrix P' in proposed method 1 and proposed method 2. Therefore, first, the method for calculating the cross-disease predicted adjacency matrix P' in the conventional method will be described.

[0154]

number

[0155] Formula 9 is a calculation formula for the cross-disease prediction adjacency matrix P' in the conventional method. q and v are hyperparameters that take any value between 0 and 1. q is a cross-disease treatment similarity weight that adjusts the contribution rate of the cross-disease treatment similarity matrix T'. v is a cross-disease biomarker similarity weight that adjusts the contribution rate of the cross-disease biomarker similarity matrix B'.

[0156] To compare the prediction accuracy of proposed method 1, proposed method 2, and the conventional method, we performed leave-one-out cross-validation for each method. Leave-one-out cross-validation is a validation method in which one sample is removed from the dataset as a test sample, and predictions are evaluated for the test sample using the remaining data. This process is repeated for all samples in the dataset.

[0157] As data for leave-one-out cross-validation, we used data of 824 sets of known ternary relationships that were clinically and experimentally confirmed for treatment entities, biomarker entities, and disease entities contained in the public database TheMarker (https: / / themarker.idrblab.cn). In each trial of leave-one-out cross-validation in this example, we used data of 824 sets of known ternary relationships that were clinically and experimentally confirmed for treatment entities, biomarker entities, and disease entities. j , biomarker entity b k , and disease entity d l The known ternary relation (t j , b k , d l ) was deleted from the 824 pairs of ternary relation data 300, and the cross-disease predicted adjacency matrix P' was calculated from the remaining 823 pairs of data. j , b k ) prediction scores were evaluated.

[0158] Usually, in leave-one-out cross-validation, the data used as test samples, i.e., the data to be removed, are all samples in the dataset. However, for a certain treatment entity t j, biomarker entity b k , and disease entity d l The known ternary relation (t j , b k , d l ) is deleted, the binary relation (t j , b k ) becomes completely unpredictable, or the binary relation (t j , b k ) prediction is sometimes unnecessary. Therefore, the corresponding ternary relations (t j , b k , d l ) were excluded from the test samples for leave-one-out cross-validation in this example. The exclusion criteria are described below.

[0159] A treatment entity t j , biomarker entity b k , and disease entity d l The known ternary relation (t j , b k , d l ) for t j or b k does not exist in the remaining 823 pairs of data, we use the ternary relation (t j , b k , d l ) was excluded. This is because the ternary relation (t j , b k , d l ) is deleted, the binary relation (t j , b k ) is completely impossible to predict.

[0160] Also, a certain treatment entity t j , biomarker entity b k , and disease entity d l The known ternary relation (t j , b k , d l ) for d l The known ternary relationship between different diseases (tj , b k , dm) exists in the remaining 823 pairs of data, the ternary relation (t j , b k , d l ) was excluded. This is because the ternary relation (t j , b k , d l ) is deleted, the remaining 823 pairs of data contain binary relations (t j , b k ) exists.

[0161] After applying the above exclusion criteria, 375 ternary relations remained to be tested in leave-one-out cross-validation.

[0162] The evaluation indices for prediction accuracy in leave-one-out cross-validation were HR@N (Hit Rate at N) and detection rate.

[0163] HR@N represents the percentage of test samples deleted by leave-one-out cross-validation that are in the top N in the ranking of cross-disease prediction scores for binary relationships between therapeutic entities and biomarker entities. Note that only unknown binary relationships are included in the ranking, and known binary relationships are excluded. For example, the statement "HR@10 is 50%" means that "50% (187 out of 375 pairs) of the test samples evaluated by leave-one-out cross-validation have prediction scores in the top 10." A high HR@N indicates a low number of predicted false edges (binary relationships), i.e., high prediction specificity.

[0164] The detection rate represents the proportion of test samples in leave-one-out cross-validation that have a cross-disease prediction score greater than 0. For example, the statement "detection rate is 50%" means that "50% (187 out of 375 pairs) of the test samples evaluated by leave-one-out cross-validation have a prediction score greater than 0." A high detection rate represents high prediction sensitivity.

[0165] In the leave-one-out cross-validation, the hyperparameter values ​​were set as p = q = u = v = 0.25 in the calculation of the disease-specific predicted adjacency matrix (Equation 5) by Proposed Method 1 and Proposed Method 2. In addition, the hyperparameter values ​​were set as q = v = 0.5 in the calculation of the cross-disease predicted adjacency matrix (Equation 9) by the conventional method.

[0166] Figure 14 shows the HR@N results for proposed method 1, proposed method 2, and the conventional method. The horizontal axis represents the ranking, and the vertical axis represents the HR (Hit Rate). For example, data point 1401 indicates that the HR@10 is 24.5% when the conventional method is used.

[0167] As shown in Figure 14, overall, the HR@N when using proposed method 1 and the HR@N when using proposed method 2 are roughly equal, and both tend to be higher than the HR@N when using the conventional method. This suggests that proposed methods 1 and 2 have higher prediction specificity than the conventional method.

[0168] In the area 1402 enclosed by the dotted line in Figure 14, the HR rises sharply relative to the change in rank. This indicates that the cross-disease prediction score of the test sample is 0, i.e., the test sample is at the bottom. The detection rate using proposed method 1 is 73.3%, and the detection rates using proposed method 2 and the conventional method are 73.6%. The detection rate using proposed method 2 is 0.3% higher than the detection rate using proposed method 1, suggesting that proposed method 2 has higher prediction sensitivity than proposed method 1.

[0169] As described above, the information processing system of Example 1 or Example 2 can accurately predict an unknown binary relationship between a treatment method and a biomarker.

[0170] In the information processing systems of Examples 1 and 2, the similarity between nodes is calculated based on the number of commonly connected adjacent treatment nodes (see Formulas 1, 2, 7, and 8). However, other similarity indices such as SimRank or Jaccard index, or similarities based on embedding representations of nodes calculated by machine learning techniques such as Node2Vec or graph neural networks, may be used instead, or these may be combined. This can improve the prediction accuracy of unknown binary relationships between treatments and biomarkers.

[0171] The present invention can be extended to a four- or more-term relationship by adding other types of entities in addition to treatment, biomarker, and disease. For example, a four-term relationship between treatment, biomarker, disease, and side effect becomes a three-term relationship between treatment, biomarker, and the combination of disease and side effect by considering the combination of disease and side effect as a single type of entity. In this way, by converting a four- or more-term relationship into a three-term relationship, the present invention can be extended to a four- or more-term relationship.

[0172] 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.

[0173] 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.

[0174] 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.

[0175] In addition, the control lines and information lines shown are those that are considered necessary for explanation, and do not necessarily represent all the control lines and information lines that are necessary for implementation. In reality, it can be assumed that almost all components are interconnected. [Explanation of symbols]

[0176] 101 Server 102 databases 103 Input Device 104 Output Device 105 Arithmetic equipment 106 memory 107 Storage device 108 Bipartite graph generation unit 109 Node Similarity Calculation Unit 110 Unknown edge prediction unit 111 Output section 112 Entity Set Data Storage 113 Known ternary relation data storage unit

Claims

1. An information processing system for predicting an unknown binary relationship between a treatment and a biomarker from a known ternary relationship between a treatment, a biomarker, and a disease, comprising: The computer is configured by an arithmetic unit that executes predetermined processing and a storage device connected to the arithmetic unit, a bipartite graph generation unit configured to generate, for each disease, a disease-specific bipartite graph that expresses the binary relationship between the treatment and the biomarker from the known ternary relationship; the computing device is an inter-node similarity calculation unit that calculates, from the disease-specific bipartite graph, a disease-specific treatment similarity between the treatments for each disease, a cross-disease treatment similarity between the treatments for all diseases, a disease-specific biomarker similarity between the biomarkers for each disease, and a cross-disease biomarker similarity between the biomarkers for all diseases; an unknown edge prediction unit configured to calculate at least one of a disease-specific prediction score and a cross-disease prediction score of an unknown edge using the disease-specific bipartite graph, the disease-specific treatment similarity, the cross-disease treatment similarity, the disease-specific biomarker similarity, and the cross-disease biomarker similarity; an output unit that outputs at least one of the disease-specific treatment similarity, the cross-disease treatment similarity, the disease-specific biomarker similarity, the cross-disease biomarker similarity, the disease-specific prediction score, and the cross-disease prediction score.

2. 2. The information processing system according to claim 1, The bipartite graph generation unit generates a disease-specific adjacency matrix G that represents the edges of the disease-specific bipartite graph. (i) Generate The inter-node similarity calculation unit The disease-specific adjacency matrix G (i) Based on this, a similarity matrix T (i) , The disease-specific adjacency matrix G (i) Based on the above, a disease-specific biomarker similarity matrix B representing the similarity between the disease-specific biomarkers is (i) , The disease-specific adjacency matrix G (i) and the disease-specific treatment method similarity matrix T (i) Based on the above, a cross-disease treatment similarity matrix T' representing the cross-disease treatment similarity, and The disease-specific adjacency matrix G (i) and the disease-specific biomarker similarity matrix B (i) and calculating a cross-disease biomarker similarity matrix B' representing the cross-disease biomarker similarity based on the above formula.

3. 3. The information processing system according to claim 2, The inter-node similarity calculation unit The disease-specific adjacency matrix G (i) Based on the commonality of nodes in the disease-specific treatment similarity matrix T (i) , The disease-specific adjacency matrix G (i) Based on the commonality of the nodes in the disease-specific biomarker similarity matrix B (i) , The disease-specific adjacency matrix G (i) and the disease-specific treatment method similarity matrix T (i) the cross-disease treatment similarity matrix T' based on the commonality of nodes in The disease-specific adjacency matrix G (i) and the disease-specific biomarker similarity matrix B (i) The information processing system is characterized by calculating the cross-disease biomarker similarity matrix B' based on the commonality of nodes in the

4. 4. The information processing system according to claim 3, The inter-node similarity calculation unit The disease-specific adjacency matrix G (i) The more commonly connected nodes there are in the disease-specific treatment similarity matrix T (i) , and The disease-specific adjacency matrix G (i) The more commonly connected nodes there are in the disease-specific biomarker similarity matrix B, the higher the similarity. (i) , The disease-specific adjacency matrix G (i) and the disease-specific treatment method similarity matrix T (i) The cross-disease treatment similarity matrix T', in which the similarity increases as the number of nodes in The disease-specific adjacency matrix G (i) and the disease-specific biomarker similarity matrix B (i) The information processing system is characterized by calculating the cross-disease biomarker similarity matrix B' in which the similarity increases as the number of nodes in the matrix increases.

5. 2. The information processing system according to claim 1, the bipartite graph generation unit extracts known binary relations between the treatments and the biomarkers from the known ternary relations, and generates a cross-disease bipartite graph representing the binary relations between the treatments and the biomarkers and a cross-disease adjacency matrix G' representing edges of the cross-disease bipartite graph; the node similarity calculation unit calculates a cross-pathology therapy similarity matrix T' and a cross-pathology biomarker similarity matrix B' based on the commonality of nodes in the cross-pathology adjacency matrix G'.

6. 3. The information processing system according to claim 2, The unknown edge prediction unit calculates the disease-specific adjacency matrix G (i) , the disease-specific treatment method similarity matrix T (i) , the disease-specific biomarker similarity matrix B (i) , the cross-disease inter-treatment similarity matrix T′, and the cross-disease inter-biomarker similarity matrix B′ are used to generate a disease-specific prediction adjacency matrix P (i) and a cross-disease prediction adjacency matrix P' representing the cross-disease prediction score.

7. 7. The information processing system according to claim 6, The unknown edge prediction unit calculates the disease-specific treatment similarity matrix T (i) and the cross-disease treatment similarity matrix T' to the disease-specific adjacency matrix G (i) and the disease-specific biomarker similarity matrix B (i) and the cross-disease biomarker similarity matrix B' to the disease-specific adjacency matrix G (i) The disease-specific prediction adjacency matrix P (i) and an information processing system for calculating the cross-disease prediction adjacency matrix P'.

8. 2. The information processing system according to claim 1, The information processing system is characterized in that the output unit outputs a binary relationship between the treatment method and the biomarker that has a similarity greater than a predetermined threshold and a prediction score greater than a predetermined threshold.

9. 2. The information processing system according to claim 1, The output unit Accepts as input the treatment to be displayed; extracting a treatment method whose similarity to the input treatment method is equal to or greater than a predetermined threshold, and outputting the extracted treatment method; or An information processing system characterized by extracting biomarkers whose prediction scores with the input treatment methods are equal to or greater than a predetermined threshold, and outputting the extracted biomarkers.

10. 2. The information processing system according to claim 1, The output unit Accepts as input the biomarkers to be displayed; extracting biomarkers whose similarity to the input biomarkers is equal to or greater than a predetermined threshold, and outputting the extracted biomarkers; or An information processing system characterized by extracting treatments whose prediction scores with the input biomarkers are equal to or greater than a predetermined threshold, and outputting the extracted treatments.

11. 2. The information processing system according to claim 1, The information processing system is characterized in that the output unit displays the disease-specific treatment similarity, the cross-disease treatment similarity, the disease-specific biomarker similarity, the cross-disease biomarker similarity, the disease-specific prediction score, or the cross-disease prediction score in a table format.

12. 2. The information processing system according to claim 1, The information processing system is characterized in that the output unit displays the known binary relationship between the treatment and the biomarker, and the predicted binary relationship between the treatment and the biomarker, in at least one of the disease-specific bipartite graph and the cross-disease bipartite graph for the treatment and the biomarker.

13. A prediction method in which an information processing system predicts an unknown binary relationship between a treatment and a biomarker from a known ternary relationship between a treatment, a biomarker, and a disease, comprising: the information processing system is configured by a computer having an arithmetic unit that executes predetermined processing and a storage device connected to the arithmetic unit; The prediction method includes: a bipartite graph generation step in which the arithmetic device generates, for each of the diseases, a disease-specific bipartite graph that expresses the binary relationship between the treatment and the biomarker from the known ternary relationship; a node-to-node similarity calculation step in which the arithmetic device calculates, from the disease-specific bipartite graph, disease-specific treatment similarities between the treatments for each disease, cross-disease treatment similarities between the treatments for all diseases, disease-specific biomarker similarities between the biomarkers for each disease, and cross-disease biomarker similarities between the biomarkers for all diseases; an unknown edge prediction step in which the calculation device calculates at least one of a disease-specific prediction score and a cross-disease prediction score of an unknown edge using the disease-specific bipartite graph, the disease-specific treatment similarity, the cross-disease treatment similarity, the disease-specific biomarker similarity, and the cross-disease biomarker similarity; and an output step in which the arithmetic device outputs at least one of the disease-specific treatment similarity, the cross-disease treatment similarity, the disease-specific biomarker similarity, the cross-disease biomarker similarity, the disease-specific prediction score, and the cross-disease prediction score.

Images