Method and device for predicting combination of medicines

Abstract

To provide a method and a device for predicting a combination of medicines useful in various fields such as medical care, pharmacy, and drug discovery.SOLUTION: This method for predicting a combination of medicines comprises the steps of: building an inter-protein interaction network in which a disease module, a medicine A module, and a medicine B module are mapped in a comprehensive human PPI; acquiring a positional relation, proximity, and correlation among these modules on the inter-protein interaction network; calculating a positional relation score, a distance relation score, and a correlation score on the basis of the positional relation, the proximity, and the correlation which have been acquired; and calculating a prediction score indicating an effect when the medicine A and the medicine B are combined for a disease on the basis of the positional relation score, the proximity score, and the correlation score which have been acquired.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a drug combination prediction method and a drug combination prediction device that are useful in various fields such as medical care, pharmaceuticals, drug discovery, and biological research. [Background technology]

[0002] Combination therapy, which combines drugs, is effective for treating complex diseases such as cancer and diabetes. Combination therapy is expected to improve therapeutic effects, reduce the dosage of each drug, and reduce side effects, but drug combinations are often determined by intuition or experience, or require large-scale biological experiments. However, even if drug combinations are determined by intuition or experience, the number of drug combinations is enormous, and it is not realistic to try all combination patterns. Therefore, there is a demand for computer-based drug combination prediction, and the prior art is described in Non-Patent Documents 1 and 2.

[0003] Non-Patent Document 1 discloses a technology for predicting drug combinations based on a network. This technology uses protein-protein interaction information to predict drug combinations that will have a synergistic effect based on the positional relationship between a disease and two drugs on the interactome and the distance relationship between the drugs. In addition, Non-Patent Document 2 discloses a technique for characterizing disease and disease relationships based on a network from the perspective of drugs and therapeutic targets. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Cheng F, Kovacs IA, Barabasi AL. Network-based prediction of drug combinations. Nat Commun. 2019 Mar 13;10(1):1197. doi: 10.1038 / s41467-019-09186-x. Erratum in: Nat Commun. 2019 Apr 15;10(1):1806. PMID: 30867426; PMCID: PMC6416394. [Non-Patent Document 2] Iida M, Iwata M, Yamanishi Y. Network-based characterization of disease-disease relationships in terms of drugs and therapeutic targets. Bioinformatics. 2020 Jul 1;36(Suppl_1):i516-i524. doi: 10.1093 / bioinformatics / btaa439. PMID: 32657408; PMCID: PMC7355285. Summary of the Invention [Problem to be solved by the invention]

[0005] However, the combination prediction method disclosed in Non-Patent Document 1 has a problem of low prediction accuracy because it captures the synergistic effects of drugs from only one aspect of a life phenomenon, namely protein-protein interactions. In addition, the method for characterizing diseases and disease relationships disclosed in Non-Patent Document 2 predicts new applicable drugs using protein-protein interaction information and disease-specific gene expression profiles, but is unable to predict drug combinations.

[0006] In short, identifying synergistic drug combinations is important in drug development, but the number of drugs on the market is huge, and some of their targets are unclear, and there are many diseases for which no therapeutic drugs or treatment methods have been established. In light of this situation, the combination prediction method disclosed in Non-Patent Document 1 is an effective means for predicting drug combinations, but since it only uses information on protein-protein interactions, the prediction accuracy is low, and a new prediction method with higher prediction accuracy is needed.

[0007] Therefore, an object of the present invention is to provide a new drug combination prediction method and drug combination prediction device with high prediction accuracy. [Means for solving the problem]

[0008] The drug combination prediction method of the present invention is a protein-protein interaction network in which gene information is represented as nodes and information on the action and effect between the gene information is represented as edges connecting the nodes, (1-1) A disease module is a collection of multiple pieces of disease cause-related gene information, the disease cause-related gene information being related to the cause of a certain disease; (1-2) A drug A module, which is a collection of multiple pieces of drug A-related genetic information, and which is related to drug A; (1-3) A drug B module, which is a collection of multiple pieces of drug B-related genetic information, the genetic information being related to a drug B different from the drug A; and A step of constructing a protein-protein interaction network by mapping the above onto comprehensive human protein-protein interactions (PPIs); in protein-protein interaction networks (2-1) Disease module and drug A module (2-2) Disease module and drug B module determining a positional relationship indicating whether the disease module and the drug A module overlap in the protein-protein interaction network, whether the disease module and the drug B module overlap in the protein-protein interaction network, and whether the drug A module and the drug B module overlap in the protein-protein interaction network based on the above; determining proximity (the distance relationship between (2-1) and (2-2) in the protein-protein interaction network) indicating the likelihood of a therapeutic effect on a disease when drug A or drug B is administered alone based on (2-1) and (2-2); determining a correlation indicating the magnitude of the effect on a disease when drug A or drug B is administered alone based on (2-1) and (2-2) (correlation between (2-1) or (2-2) in the protein-protein interaction network and gene expression information of the disease); Based on the positional relationship, a positional relationship score T QAB and Based on the proximity (distance relationship), the distance relationship score P QAB and Based on the correlation, a correlation score C for the disease module, drug A module, and drug B module QAB and Positional relationship score T QAB , distance relationship score P QAB , and correlation C QAB determining a prediction score indicating the effect of combining drug A and drug B against a disease based on the score; Includes.

[0009] In addition, the drug combination prediction method includes, before the step of determining the correlation, a step of estimating new disease cause related genes, new drug A related genes, and new drug B related genes using network propagation, and determining new disease cause related genes that were not present in the disease module, new drug A related genes that were not present in the drug A module, and new drug B related genes that were not present in the drug B module. It is preferable that the present invention includes the following:

[0010] As a result, according to the drug combination prediction method of the present invention, a prediction score is obtained based on the correlation between the disease module, drug A module, and drug B module (correlation score) in addition to the positional relationship (positional relationship score) and proximity (distance relationship score) between them. In addition, by including a step of using network propagation to find new disease cause related genes, new drug A related genes, and new drug B related genes, the newly found (identified) overlapping genes can be used to find expression correlations when finding correlations (correlation scores).

[0011] The drug combination prediction device of the present invention includes a storage means for storing protein-protein interaction data, disease cause-related gene data, disease-specific gene expression data, and drug-specific gene expression data; With reference to the protein-protein interaction data, the disease cause-related gene data, and the drug-related gene information identified from the drug-specific gene expression data, This is a protein-protein interaction network in which gene information is represented as nodes and information on the action and effect between gene information is represented as edges connecting the nodes. (1-1) A disease module is a collection of multiple pieces of disease cause-related gene information, the disease cause-related gene information being related to the cause of a certain disease; (1-2) A drug A module, which is a collection of multiple pieces of drug A-related genetic information, and which is related to drug A; (1-3) A drug B module, which is a collection of multiple pieces of drug B-related genetic information, the genetic information being related to a drug B different from the drug A; and A first calculation means for constructing a protein-protein interaction network by mapping the above onto comprehensive human protein-protein interactions (PPIs); in protein-protein interaction networks (2-1) Disease module and drug A module (2-2) Disease module and drug B module Based on the above, a positional relationship indicating whether the disease module and the drug A module overlap in the protein-protein interaction network, whether the disease module and the drug B module overlap in the protein-protein interaction network, and whether the drug A module and the drug B module overlap in the protein-protein interaction network is displayed. Based on (2-1) and (2-2), the proximity (the distance relationship between (2-1) and (2-2) in the protein-protein interaction network) indicates the likelihood of a therapeutic effect on a disease when drug A or drug B is administered alone. With reference to the disease-specific gene expression data and the drug-specific gene expression data, a correlation indicating the magnitude of the effect on the disease when drug A or drug B is administered alone based on (2-1) and (2-2) (correlation between (2-1) or (2-2) in the protein-protein interaction network and gene expression information of the disease) is determined. A second calculation means for calculating Based on the positional relationship, a positional relationship score T QAB and, Based on the proximity (distance relationship), the distance relationship score P QAB and, Based on the correlation, a correlation score C for the disease module, drug A module, and drug B module QAB and, A third calculation means for calculating Positional relationship score T QAB , distance relationship score P QAB , and the correlation score C QAB A fourth calculation means for calculating a prediction score indicating an effect of a combination of drug A and drug B on a disease based on the above; has.

[0012] In addition, in the drug combination prediction device, the storage means further stores disease similarity data and drug similarity data, The second calculation means refers to the inter-disease similarity data and the inter-drug similarity data, and uses network propagation to estimate new disease cause related genes, new drug A related genes, and new drug B related genes, to obtain new disease cause related genes that were not present in the disease module, new drug A related genes that were not present in the drug A module, and new drug B related genes that were not present in the drug B module, and then calculates correlations. Effect of the Invention

[0013] According to the drug combination prediction method of the present invention, a prediction score is obtained based on the correlation between the disease module, drug A module, and drug B module (correlation score) in addition to the positional relationship (positional relationship score) and proximity (distance relationship score) between them, making it possible to predict drug combinations with greater accuracy than conventional methods.

[0014] Furthermore, the drug combination prediction device of the present invention can achieve the same effects as the drug combination prediction method of the present invention. [Brief description of the drawings]

[0015] [Figure 1] FIG. 1 is a schematic flow diagram of a drug combination prediction method according to the present invention. [Diagram 2] 1 is a schematic functional block diagram of a drug combination prediction device according to the present invention. [Diagram 3](A) is a schematic diagram of the protein-protein interaction network to be constructed, and (B) is a schematic diagram for explaining the positional relationship of the disease module, drug A module, and drug B module in the protein-protein interaction network. [Figure 4] (A) is a schematic diagram for explaining the proximity of a disease module and a drug A module in a protein-protein interaction network, and (B) is a schematic diagram for explaining the proximity of a disease module and a drug B module in a protein-protein interaction network. [Diagram 5] (A) is a schematic diagram for explaining the correlation between a disease module and a drug A module in a protein-protein interaction network, and (B) is a schematic diagram for explaining the correlation between a disease module and a drug B module in a protein-protein interaction network. [Figure 6] FIG. 1 is a diagram for explaining the process of finding new disease causal related genes, etc., using network propagation; (A) is a schematic diagram showing the state before propagation, and (B) is a schematic diagram showing the state after propagation. [Figure 7] This figure explains the process of calculating the positional relationship scores of the disease module, drug A module, and drug B module, where (A) is a schematic diagram showing a positional relationship classified as class 1, and (B) is a schematic diagram showing a positional relationship classified as class 2. [Figure 8] This figure explains the process of calculating the positional relationship scores of the disease module, drug A module, and drug B module, where (A) is a schematic diagram showing a positional relationship classified as class 3-1, and (B) is a schematic diagram showing a positional relationship classified as class 3-2. [Figure 9] This figure explains the process of calculating the positional relationship scores of the disease module, drug A module, and drug B module, where (A) is a schematic diagram showing a positional relationship classified as class 4-1, and (B) is a schematic diagram showing a positional relationship classified as class 4-2. [Figure 10]This figure explains the process of calculating the positional relationship scores of the disease module, drug A module, and drug B module, where (A) is a schematic diagram showing a positional relationship classified into class 5, and (B) is a schematic diagram showing a positional relationship classified into class 6. [Figure 11] FIG. 1 shows the results of an experiment using the drug combination prediction method according to the present invention, in which (A) is a schematic diagram showing the number of genes whose expression was significantly altered (DEGs), and (B) is a diagram showing the amount of change in DEGs. [Figure 12] FIG. 1 shows the results of an experiment using the drug combination prediction method of the present invention, illustrating the effect of the mode of action of combined exposure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] The following describes an embodiment of the present invention in detail. The description of the components described below is an example (representative example) of an embodiment of the present invention, and the present invention is not limited to the following content unless the gist of the present invention is changed.

[0017] [Drug combination prediction device] The drug combination prediction device 10 has a communication means 11, a storage means 12, a display means 13, an input means 14, a first calculation means 15, a second calculation means 16, a third calculation means 17, and a fourth calculation means 18 (see FIG. 2).

[0018] [Means of communication] The communication means 11 is connected to an external network by wire or wirelessly, and has a function of communicating with various terminals (PCs, smartphones, tablets) and various databases via the network.

[0019] [Storage means] The storage means 12 is a memory, a database, etc. that stores various data. The storage means 12 stores the following data: (1) Protein-protein interaction data (2) Disease-related genetic data (3) Disease-specific gene expression data (4) Drug-specific gene expression data (5) Similarity data between diseases, similarity data between drugs (6) Computational data for constructing protein-protein interaction networks (7) Computational data for identifying new disease-related genes using network propagation (8) Calculation data for calculating location scores, distance scores, correlation scores, prediction scores, etc. (9) Other data necessary for control, various programs for operating the drug combination prediction device, etc.

[0020] In particular, (1) protein-protein interaction data to (4) drug-specific gene expression data can be obtained from an external database via communication means 11, or imported from a USB or an external hard disk, and stored in memory means 12.

[0021] For example, protein-protein interaction data can be obtained from databases such as HuRI (Human Reference Protein Interactome Mapping), CORUM (Comprehensive Resource of Mammalian protein complexes), PhosphoSitePlus, KEGG (Kyoto encyclopedia of genes and genomes), InnateDB, INstruct, and SignaLink.

[0022] On the other hand, disease causation-related gene data can be obtained from databases such as OMIM (Online Mendelian Inheritance in Man), ClinVar, DisGeNET, PheWAS (Phenome Wide Association Studies), and GWAS (Genome Wide Association Study).

[0023] In addition, disease-specific gene expression data can be obtained from databases such as GEO (Gene Expression Omnibus) and CREEDS (Crowd Extracted Expression of Differential Signatures). Drug-specific gene expression data can be obtained from the LINCS (The Library of Integrated Network-Based Cellular Signatures) database.

[0024] In addition, the disease similarity data can be calculated from the distance in the protein-protein interaction network between diseases based on the data obtained from the CREEDS database. Specifically, the calculation can be performed using a method developed by the present inventor (see Non-Patent Document 2). In addition, drug similarity data can be calculated based on the chemical structure of drug data obtained from the LINCS database.

[0025] Of course, various data used in the drug combination prediction method of the present invention can be obtained from databases other than the above-mentioned databases, whether or not they are paid for and free of charge.

[0026] [Display means] The display means 13 displays various information such as the prediction score, positional relationship score, distance relationship score, correlation score, or intermediate calculation data obtained by the drug combination prediction device 10 on a display or a user's terminal.

[0027] [Input method] The input means 14 is, for example, a keyboard, a touch panel, etc. The input means 14 is used to give control instructions from outside the drug combination prediction device 10 using a command line or the like, and to input various input information such as information on protein-protein interactions and gene expression information.

[0028] [Calculation means] The first calculation means 15 to the fourth calculation means 18 perform calculations required for predicting drug combinations (each step shown in FIG. 1). For convenience, in this embodiment, the roles of the first calculation means 15 to the fourth calculation means 18 are divided as follows: [1] First calculation means 15 A calculation is performed to construct a protein-protein interaction network (step S10). [2] Second calculation means 16 A positional relationship is obtained as a relationship between a disease and a drug based on a protein-protein interaction network (step S21). In addition, proximity is calculated as a relationship between a disease and a drug based on the protein-protein interaction network (step S22). In addition, a correlation is obtained as a relationship between a disease and a drug based on the correlation of gene expression levels (step S24). At this time, if necessary, a new drug-related gene is obtained using network propagation before the correlation is obtained (step S23). [3] Third calculation means 17 Based on the positional relationship, proximity, and correlation determined by the second calculation means 16, a positional relationship score, a distance relationship score, and a correlation relationship score are determined (step S30). [4] Fourth calculation means 18 A predicted score is calculated based on the positional relationship score, the distance relationship score, and the correlation relationship score calculated by the third calculation means 17 (step S40).

[0029] [Drug combination prediction method] Next, as an embodiment of the present invention, an example will be described in which the drug combination prediction device 10 shown in FIG. 2 executes a drug combination prediction method in accordance with the schematic flow shown in FIG.

[0030] (1. Constructing protein-protein interaction networks) First, the drug combination prediction device 10 constructs, by the first calculation means 15, a protein-protein interaction network (see FIG. 3(A)) in which genetic information is represented as nodes and information regarding the action and effect between the genetic information is represented as edges connecting the nodes (step S10). Specifically, the first calculation means 15 constructs a protein-protein interaction network in which a disease module, a drug A module, a drug B module, and a drug B module are mapped onto comprehensive human PPIs.

[0031] At this time, the first calculation means 15 constructs a protein-protein interaction network by referring to the protein-protein interaction data stored in the memory means 12, the disease cause-related gene data, and the drug-related gene information identified from the drug-specific gene expression data.

[0032] The disease module referred to here is disease cause-related gene information in which gene information is related to the cause of a certain disease, and is a collection of multiple disease cause-related gene information. The Drug A module here is Drug A-related genetic information whose genetic information is related to Drug A, and is a collection of multiple pieces of Drug A-related genetic information. On the other hand, the Drug B module here is Drug B-related genetic information whose genetic information is related to Drug B different from Drug A, and is a collection of multiple pieces of Drug B-related genetic information. In this description, the drug A module and the drug B module may be collectively referred to as the "drug module," and the drug A-related gene information and the drug B-related gene information may be collectively referred to as the "drug-related gene information."

[0033] (2-1. Determine the positional relationship) Then, the drug combination prediction device 10 calculates, by the second calculation means 16, a positional relationship as a relationship between the disease and the drug based on the protein-protein interaction network (step S21).

[0034] Specifically, the second calculation means 16 calculates the protein-protein interaction network constructed by the above-mentioned steps. (a) Disease module and drug A module (b) Disease module and drug B module Based on this, a positional relationship is identified that indicates whether the disease module and the drug A module overlap in the protein-protein interaction network, whether the disease module and the drug B module overlap in the protein-protein interaction network, and whether the drug A module and the drug B module overlap in the protein-protein interaction network.

[0035] For example, in the example shown in Figure 3(B), in the protein-protein interaction network, the disease module and the drug A module overlap, and the disease module and the drug B module also overlap, but the drug A module and the drug B module do not overlap.

[0036] (2-2. Seeking Proximity) Furthermore, the drug combination prediction device 10 calculates proximity as a relationship between a disease and a drug based on the protein-protein interaction network by the second calculation means 16 (step S22).

[0037] Specifically, the second calculation means 16 determines a proximity indicating the likelihood of a therapeutic effect on a disease when drug A or drug B is administered alone, based on the (a) and (b). This proximity is determined as a distance relationship between the (a) and (b) in the protein-protein interaction network (distance on the protein-protein interaction network).

[0038] For example, Figure 4(A) is a schematic diagram showing the distance relationship between a disease module and a drug A module in a protein-protein interaction network, and Figure 4(B) is a schematic diagram showing the distance relationship between a disease module and a drug B module in the protein-protein interaction network.In this way, the degree to which each module is distant (close) in the protein-protein interaction network is calculated as the proximity.

[0039] (2-3. Finding correlations) Furthermore, the drug combination prediction device 10 calculates a correlation as a relationship between a disease and a drug based on the correlation of gene expression levels by the second calculation means 16 (step S24).

[0040] Specifically, the second calculation means 16 determines a correlation indicating the magnitude of the effect on the disease when the drug A or the drug B is administered alone, based on the (a) and (b). This correlation is determined as a correlation between the (a) or (b) and gene expression information of the disease in the protein-protein interaction network.

[0041] For example, the schematic diagram shown in Figure 5(A) shows a process of calculating the expression correlation between the disease to be treated and drug A, i.e., the magnitude of the effect on the disease to be treated when drug A is administered alone, using the expression level in human cells when drug A is administered. In this case, as shown in Figure 5(A), if there is an overlapping portion (overlapping genes) between the disease module and the drug A module, it is possible to calculate the expression correlation between the disease to be treated and drug A, but as shown in Figure 5(B), if there is no overlapping portion (overlapping genes) between the disease module and the drug B module, it is not possible to calculate the expression correlation between the disease to be treated and drug B.

[0042] (2-4. Searching for new drug-related genes) Therefore, in such a case, the second calculation means 16 can obtain new drug-related genes using network propagation before obtaining the correlation (step S23). Such a case is a case where there are no or few overlapping genes belonging to a module of a disease to be treated and a module of a drug.

[0043] Specifically, the second calculation means 16 uses network propagation to increase genes belonging to a certain module (disease module or drug module). Figures 6(A) and (B) are schematic diagrams showing genes of a certain module before and after propagation, and as shown in Figure 6(B), new genes (new disease-related genes when a certain module is a disease module, and new drug-related genes when a certain module is a drug module) are increased. This allows genes that overlap between the disease module and the drug module to be identified and expression correlations to be calculated.

[0044] In this embodiment, the network propagation method used is the algorithm described in Reference 1. Reference 1 discloses a technique for associating genes and protein complexes with diseases by network propagation (PRINCE).

[0045] (Reference 1) Vanunu, O., Magger, O., Ruppin, E., Shlomi, T., & Sharan, R. (2010). Associating genes and protein complexes with disease via network propagation. PLoS Computational Biology, 6(1), e1000641. https: / / doi.org / 10.1371 / journal.pcbi.1000641

[0046] When this algorithm is used, the disease similarity data and drug similarity data listed as one of the data stored in the storage means 12 are used to propagate disease-causing gene information and drug-related gene information. Of course, other algorithms may be used as the network propagation method.

[0047] (3-1. Calculate the positional score) Next, the drug combination prediction device 10 calculates the positional relationship scores T QAB is calculated (step S30).

[0048] Specifically, the third calculation means 17 classifies the disease module, the drug A module, and the drug B module identified by the second calculation means 16 into one of the following classes 1 to 6 to which a score is assigned based on the positional relationship, thereby obtaining a positional relationship score T QAB Request.

[0049] (Class 1) The drug A module, the drug B module, and the disease module all overlap (see FIG. 7(A)). (Class 2) The drug A module and the disease module overlap, and the drug B module and the disease module also overlap, but the drug A module and the drug B module do not overlap (see Figure 7(B)). (Class 3-1) The drug A module and the drug B module overlap, and the drug A module and the disease module also overlap, but the drug B module and the disease module do not overlap (see Figure 8(A)). (Class 3-2) The drug A module and the drug B module overlap, and the drug B module and the disease module also overlap, but the drug A module and the disease module do not overlap (see Figure 8(B)). (Class 4-1) The drug A module and the disease module overlap, but the drug B module does not overlap with the drug A module and the disease module (see Figure 9(A)). (Class 4-2) The drug B module and the disease module overlap, but the drug A module does not overlap with the drug B module or the disease module (see Figure 9(B)). (Class 5) The drug A module and the drug B module overlap, but the disease module does not overlap with the drug A module and the drug B module (see Figure 10(A)). (Class 6) The drug A module, the drug B module, and the disease module do not overlap with each other (see FIG. 10(B)).

[0050] The above classes 1 to 6 are then assigned a positional relationship score T QAB is assigned. (Class 1):T QAB =4 (Class 2):T QAB =5 (Class 3-1, 3-2): T QAB =3 (Class 4-1, 4-2): T QAB =2 (Class 5): T QAB =0 (Class 6): T QAB =1

[0051] In other words, the example shown in Figure 3(B) is classified into class 2 and has a positional relationship score of T QAB is evaluated as “5”.

[0052] (3-2. Calculate the distance relationship score) In addition, the drug combination prediction device 10 calculates a distance relationship score P based on the proximity (distance relationship) between the disease module, the drug A module, and the drug B module obtained by the second calculation means 16. QAB is calculated (step S30).

[0053] Specifically, this proximity is evaluated using a distance measure of the protein-protein interaction network, and is found by calculating the path length between the disease-causing genes that constitute the disease module and the drug-related gene information that constitutes the drug module.

[0054] For example, as shown in the following formula (1), the proximity (distance relationship) d(Q, A) between the disease module and the drug A module is defined using the genetic information (set of genetic information) of the disease module, the genetic information (set of genetic information) of the drug A module, and the shortest path length between these genes (nodes).

[0055]

number

[0056] However, if the number of genes belonging to the drug A module is different from the number of genes belonging to the drug B module, the proximity (a) cannot be compared with the proximity (b), so the proximity d(Q, A) calculated by the above formula (1) is normalized.

[0057] In addition, the normalized distance relationship is appropriately inverted and scaled to obtain the proximity P QA Request. In addition, the proximity of the disease module and the drug B module P QB Also, proximity P QA It is calculated in the same manner as above.

[0058] And the required proximity P QB and proximity P QA The average of these is the distance relationship score P QAB It is calculated as follows.

[0059] (3-3. Calculate the correlation score) In addition, the drug combination prediction device 10 calculates a correlation score C based on the correlation between the disease module, the drug A module, and the drug B module obtained by the second calculation means 16 using the third calculation means 17. QAB is calculated (step S30).

[0060] Specifically, the correlation between the disease module and the drug A module is calculated by considering multiple disease cause-related genes included in the disease module and multiple drug A-related genes included in the drug A module as vector elements and calculating the cosine similarity C QA It can be calculated as:

[0061] Similarly, the correlation between the disease module and the drug B module can be calculated by considering multiple disease cause-related genes included in the disease module and multiple drug B-related genes included in the drug B module as vector elements and calculating the cosine similarity CQB It can be calculated as:

[0062] And the calculated cosine similarity C QA and the cosine similarity C QB The average of the absolute values ​​of QAB It is calculated as follows.

[0063] (4. Calculate the predicted score) Finally, the drug combination prediction device 10 calculates the positional relationship score T QAB , distance relationship score P QAB , and the correlation score C QAB Based on this, a prediction score indicating the effect of combining drug A and drug B against a disease is calculated (step S40).

[0064] For example, the distance relationship score P QAB and correlation score C QAB The average of these is the positional relationship score T QAB The sum of these is calculated as the prediction score.

[0065] As described above, the drug combination prediction method using the drug combination prediction device has been described as an embodiment of the present invention, but this embodiment can be modified as appropriate without departing from the gist of the present invention. For example, the second calculation means 16 calculates the positional relationship (step S21), and then calculates the positional relationship score T QAB Similarly, the second calculation means 16 may calculate the proximity and correlation (steps S22 and S23), and the second calculation means 16 may calculate the distance relationship score P QAB or correlation score C QAB may be calculated (step S30).

[0066] In addition, the positional relationship scores T QAB How to assign distance relationship score P QAB Sign reversal and scaling when calculating the correlation score C QABCalculation method and formula for predicted score (positional relationship score T QAB , distance relationship score P QAB , and the correlation score C QAB The weighting of the above can also be changed as appropriate.

[0067] [Experimental Example] Hereinafter, experimental examples of the drug combination prediction method of the present invention will be described with reference to Tables 1 to 6 and FIGS. In order to verify the effectiveness of the drug combination prediction method of the present invention (hereinafter, also referred to as the "proposed method"), the following six diseases were selected from the disease causative gene database as diseases to be treated. [1] AML (Acute myeloid leukemia) [2]CML (Chronic myeloid leukemia) [3] CC (colorectal cancer) [4]asthma [5]T2D (type 2 diabetes) [6] HTN (hypertension)

[0068] (1. Comparison of AUC) We calculated the area under curve (AUC) for each of the six diseases using the conventional method and the proposed method. The results are shown in Table 1.

[0069] The vertical axis of Table 1 indicates each method including the proposed method. The conventional method is a method for predicting drug combinations based on the proximity (distance relationship) of a disease module, a drug A module, and a drug B module described in Non-Patent Document 1. The baseline method is the proposed method without using network propagation. The horizontal axis of Table 1 indicates each disease. The numbers in parentheses indicate the number of drug combinations that have been confirmed to have a synergistic effect. Table 1 also shows the maximum AUC for each disease using each method.

[0070] [Table 1]

[0071] AUC is an index that indicates that the larger the value, the more the drug is utilized in the body. As shown in Table 1, the proposed method has the highest value for every disease compared to the conventional method and the baseline method, which shows that the proposed method can predict drug combinations with greater accuracy than the conventional method.

[0072] (2. Prediction score) In addition, focusing on CML as the disease to be treated, we used the proposed method to predict drug combinations for CML. The drug combination prediction scores calculated by the proposed method are shown in Table 2.

[0073] [Table 2]

[0074] Table 2 shows the top 20 drug combinations with the highest prediction scores. As shown in Table 2, the combination of capsaicin and mitoxantrone was predicted to have the most synergistic effect. Mitoxantrone is an anticancer drug for AML. However, according to the proposed method, it can be seen that mitoxantrone is also predicted to be effective against CML when combined with capsaicin. In other words, it is extremely difficult to determine whether a drug that is effective against one disease will also be effective against other diseases when combined with another drug, as there is no clear evidence. However, the proposed method makes it possible to evaluate drug combinations with synergistic effects by attaching an easy-to-understand index (prediction score) based on the positional relationship, proximity, and correlation between disease modules and drug modules.

[0075] In addition, the combination of mitoxantrone and topotecan (ranked 5) has been reported to be effective for CML patients (see Reference 2). This also shows that the proposed method is able to predict synergistic drug combinations to rank highly.

[0076] (Reference 2) YS Han 1, BH Kim, TH Kim, SH Dong, HJ Kim, YW Chang, JI Lee, Rin-Chang, YW Kim, JC Park, "Autoimmune hepatitis in a patient with myasthenia gravis and thymoma--a report on the first case in Korea", Korean J Intern Med . 2000 Jul;15(2):151-5. doi: 10.3904 / kjim.2000.15.2.151.

[0077] (3. In vitro experiments) Next, we performed in vitro experiments on each drug combination for CML predicted by the proposed method. The experimental results are shown in Table 3. The drug combinations shown in Table 3 are the same as those shown in Table 2 (top 20 pairs of drug combinations with high prediction scores are shown).

[0078] [Table 3]

[0079] Table 3 shows the survival rate for each drug combination: drug A, drug B, and the combination of drugs A and B. "-" indicates a combination for which the survival rate was not measured because the side effects of the drugs may outweigh the benefits, making it difficult to use in clinical practice. Additionally, to evaluate the compound effects of drugs, scores (CA score, IA score) were calculated based on the compound effect prediction models Concentration Addition (CA) and Independent Action (IA). When CA score<1, the combination is judged to have a synergistic effect, so combinations ranked 2, 9, 12, and 13 are judged to have no synergistic effect. On the other hand, other combinations can be said to have a synergistic effect, so the proposed method can be said to correctly predict many drug combinations with synergistic effects.

[0080] (4. Microarray Experiments) Furthermore, we conducted a microarray experiment focusing on the combination of capsaicin and mitoxantrone (ranked 1) as a drug combination for CML predicted using the proposed method. Specifically, through the microarray experiment, we identified a group of genes whose expression changes only with the synergistic drug combination (capsaicin and mitoxantrone).

[0081] FIG. 11 shows the results of the experiment, where FIG. 11(A) is a schematic diagram showing the number of genes whose expression was significantly altered (DEGs), and FIG. 11(B) is a diagram showing the amount of change in DEGs. As shown in FIG. 11(A), when capsaicin and mitoxantrone were combined, 1313 genes whose expression levels were altered were detected only when capsaicin and mitoxantrone were combined. Moreover, as shown in FIG. 11(B), more than half of the genes whose expression levels were altered when capsaicin and mitoxantrone were combined were genes whose expression levels were increased.

[0082] Table 4 shows the estimated pathways (biomolecular pathways) in which differentially expressed genes were enriched in each of the cases of capsaicin alone, mitoxantrone alone, and the combination of capsaicin and mitoxantrone.

[0083] [Table 4]

[0084] As shown in Table 4, the Ras1 signaling pathway is found in the pathway in only the combination (No. 2) (see underlined part). The Ras1 signaling pathway is known to play an important role in leukocyte migration and tumor progression. This logically proves that genes whose expression changes only with the synergistic drug combination of capsaicin and mitoxantrone are effective against the disease being treated (CML). Therefore, it can be said that the proposed method can correctly predict the drug combination that has a synergistic effect against the disease being treated.

[0085] Furthermore, as shown in Tables 5 and 6 and FIG. 12, it is understood that the Ras1 signaling pathway-related genes are involved in the synergistic effect of drug combinations via transcription factors (SCL: stem cell leukemia genes).

[0086] [Table 5]

[0087] [Table 6]

[0088] For example, it can be seen that the expression of genes related to the Ras1 signaling pathway is significantly increased by the combined administration (see Tables 5 and 6). Specifically, Table 5 shows the expression values ​​(Log 2 This table shows the expression levels of capsaicin and mitoxantrone in combination, which are significantly higher than those of capsaicin or mitoxantrone alone. On the other hand, Table 6 shows transcription factors enriched upstream of synergistically regulated genes (THBS1, RASGRP3, and PDGFB), and it can be seen that for all genes, the P value indicating the probability of significance is lower than the general significance level (0.05). It is also shown that transcription factors are involved in the regulation of the expression of three genes (THBS1, RASGRP3, and PDGFB) (see FIG. 12).

[0089] In other words, the two drugs (capsaicin and mitoxantrone) induce increased expression of genes related to the Ras1 signaling pathway through transcription factors, suppressing the proliferation of CML cell lines. This also shows that the proposed method can predict the appropriate combination of drugs, capsaicin and mitoxantrone, which have a synergistic effect, for the treatment of CML, which is the disease to be treated. [Industrial Applicability]

[0090] INDUSTRIAL APPLICABILITY The present invention provides a new drug combination prediction method and drug combination prediction device with high prediction accuracy, which can be used in various fields such as medical care, pharmaceuticals, drug discovery, and biological research, and is therefore industrially useful. [Explanation of symbols]

[0091] 10 Drug combination prediction device 11. Means of communication 12 Memory means 13 Display means 14 Input Methods 15 First Calculation Means 16 Second Calculation Means 17 Third calculation means 18 Fourth Calculation Means

Claims

1. a protein-protein interaction network in which gene information is represented as nodes and information on the action and effect between the gene information is represented as edges connecting the nodes, (1-1) A disease module is a collection of a plurality of pieces of disease cause-related gene information, the gene information being related to the cause of a certain disease; (1-2) a drug A module which is a collection of a plurality of pieces of drug A-related genetic information related to drug A; (1-3) The genetic information is drug B-related genetic information related to a drug B different from the drug A, and a drug B module is a collection of a plurality of the drug B-related genetic information; constructing the protein-protein interaction network by mapping the above to a comprehensive human protein-protein interaction (PPI); In the protein-protein interaction network (2-1) The disease module and the drug A module (2-2) The disease module and the drug B module determining a positional relationship indicating whether the disease module and the drug A module overlap in the protein-protein interaction network, whether the disease module and the drug B module overlap in the protein-protein interaction network, and whether the drug A module and the drug B module overlap in the protein-protein interaction network based on the above; determining a proximity (a distance relationship between (2-1) and (2-2) in the protein-protein interaction network) indicating the likelihood of a therapeutic effect on the disease occurring when the drug A or the drug B is administered alone based on (2-1) and (2-2); determining a correlation indicating the magnitude of an effect on the disease when the drug A or the drug B is administered alone based on the (2-1) and (2-2) (correlation between the (2-1) or (2-2) in the protein-protein interaction network and gene expression information of the disease); Based on the positional relationship, a positional relationship score T QAB and Based on the proximity, a distance relationship score P QAB and Based on the correlations, a correlation score C for the disease module, the drug A module, and the drug B module. QAB and The positional relationship score T QAB , the distance relationship score P QAB , and the correlation C QAB determining a prediction score indicating the effect of a combination of the drug A and the drug B on the disease based on the score; A method for predicting drug combinations comprising the steps of:

2. A step of estimating new disease cause related genes, new drug A related genes, and new drug B related genes using network propagation before the step of determining the correlation, and determining new disease cause related genes that were not present in the disease module, new drug A related genes that were not present in the drug A module, and new drug B related genes that were not present in the drug B module. The method for predicting a drug combination according to claim 1 ,

3. The positional relationship score T QAB The drug combination prediction method according to claim 1 or 2, wherein the disease module, the drug A module, and the drug B module are classified into any of the following classes 1 to 6 to which a score is assigned based on the positional relationship between the disease module, the drug A module, and the drug B module. (Class 1) The drug A module, the drug B module, and the disease module all overlap. (Class 2) The drug A module and the disease module overlap, and the drug B module and the disease module also overlap, but the drug A module and the drug B module do not overlap. (Class 3-1) The drug A module and the drug B module overlap, and the drug A module and the disease module also overlap, but the drug B module and the disease module do not overlap. (Class 3-2) The drug A module and the drug B module overlap, and the drug B module and the disease module also overlap, but the drug A module and the disease module do not overlap. (Class 4-1) The drug A module and the disease module overlap, but the drug B module does not overlap with the drug A module and the disease module. (Class 4-2) The drug B module and the disease module overlap, but the drug A module does not overlap with the drug B module and the disease module. (Class 5) The drug A module and the drug B module overlap, but the disease module does not overlap with the drug A module and the drug B module. (Class 6) The drug A module, the drug B module, and the disease module do not overlap with each other.

4. The distance relationship score P QAB based on the proximity of the disease module, the drug A module, and the drug B module, The distance relationship d(Q, A) between the disease module and the drug A module obtained by the following formula (1) is normalized and scaled to obtain the proximity P QA and, The distance relationship d(Q, B) between the disease module and the drug B module obtained by replacing drug A with drug B and using the following formula (1) is normalized and scaled to obtain the proximity P QB and, The method for predicting a drug combination according to claim 1 or 2, wherein the drug combination is determined by averaging the above.

5. The correlation score C QAB based on the correlations between the disease module, the drug A module, and the drug B module, The cosine similarity C between the plurality of disease cause related genes included in the disease module and the plurality of drug A related genes included in the drug A module is calculated by regarding them as vector elements. QA and, The cosine similarity C between the plurality of disease cause related genes included in the disease module and the plurality of drug B related genes included in the drug B module is calculated by regarding them as vector elements. QB and, 3. The method for predicting a drug combination according to claim 1 or 2, wherein the drug combination is determined by taking an average of the absolute values ​​of

6. a storage means for storing protein-protein interaction data, disease cause-related gene data, disease-specific gene expression data, and drug-specific gene expression data; With reference to the protein-protein interaction data, the disease cause-related gene data, and drug-related gene information identified from the drug-specific gene expression data, a protein-protein interaction network in which gene information is represented as nodes and information on the action and effect between the gene information is represented as edges connecting the nodes, (1-1) A disease module is a collection of a plurality of pieces of disease cause-related gene information, the gene information being related to the cause of a certain disease; (1-2) a drug A module which is a collection of a plurality of pieces of drug A-related genetic information related to drug A; (1-3) The genetic information is drug B-related genetic information related to a drug B different from the drug A, and a drug B module is a collection of a plurality of the drug B-related genetic information; a first calculation means for constructing the protein-protein interaction network by mapping the above onto a comprehensive human PPI (Protein-Protein Interaction); In the protein-protein interaction network (2-1) The disease module and the drug A module (2-2) The disease module and the drug B module Based on the above, a positional relationship indicating whether the disease module and the drug A module overlap in the protein-protein interaction network, whether the disease module and the drug B module overlap in the protein-protein interaction network, and whether the drug A module and the drug B module overlap in the protein-protein interaction network; Based on (2-1) and (2-2), a proximity indicating the likelihood of a therapeutic effect on the disease when the drug A or the drug B is administered alone (the distance relationship between (2-1) and (2-2) in the protein-protein interaction network); With reference to the disease-specific gene expression data and the drug-specific gene expression data, a correlation indicating the magnitude of the effect on the disease when the drug A or the drug B is administered alone based on (2-1) and (2-2) (a correlation between the (2-1) or (2-2) in the protein-protein interaction network and gene expression information of the disease); A second calculation means for calculating Based on the positional relationship, a positional relationship score T QAB and, Based on the proximity, a distance relationship score P QAB and, Based on the correlations, a correlation score C for the disease module, the drug A module, and the drug B module. QAB and, A third calculation means for calculating The positional relationship score T QAB , the distance relationship score P QAB , and the correlation C QAB A fourth calculation means for calculating a predicted score indicating an effect of a combination of the drug A and the drug B on the disease based on the score; A drug combination prediction device having the above-mentioned features.

7. The storage means further stores disease similarity data and drug similarity data, The second calculation means estimates the new disease cause related genes, the new drug A related genes, and the new drug B related genes by using network propagation with reference to the inter-disease similarity data and the inter-drug similarity data, and obtains the new disease cause related genes that did not exist in the disease module, the new drug A related genes that did not exist in the drug A module, and the new drug B related genes that did not exist in the drug B module, and then obtains the correlation. The drug combination prediction device according to claim 6.

Images