Drug reuse prediction method, device, storage medium and computer equipment

Abstract

The present invention relates to the field of digital medicine, and provides a drug reuse prediction method, device, storage medium and computer equipment, wherein the method includes: collecting unimodal data and multimodal association relationship data, using GCN graph neural network technology to construct a bimodal sub-graph based on the unimodal data and multimodal association relationship data, using GAT graph attention network technology based on the bimodal sub-graph to complete the feature data update for all unimodal data, and reconstructing the association relationship of the updated drug data and disease data to obtain target drug-disease association relationship data for characterizing the treatment of the target drug for the target disease. Through the present invention, the node relationship of multiple biological entities is associated, and the relationship prediction between drug characteristics and disease characteristics is obtained, which greatly improves the intelligence and accuracy of drug reuse prediction.

Description

Technical Field

[0001] The present invention relates to the field of digital medical technology, and in particular to a drug reuse prediction method, device, storage medium and computer equipment. Background Art

[0002] The pharmaceutical industry is not only one of the fastest-growing sunrise industries in the world, but also a basic and strategic industry related to the national economy and people's livelihood. Targeting the interaction between drugs and diseases is crucial in the early stages of drug discovery and reuse, and brings great hope for precision medicine and personalized treatment.

[0003] Artificial intelligence technology was first applied to pharmaceutical research and development in the medical and health field. It has been widely applied and developed in many aspects such as health management, auxiliary diagnosis and treatment, drug mining, drug preparation and even clinical rational drug use. Artificial intelligence is expected to further provide practical solutions to some challenging problems and development difficulties in the development of the pharmaceutical industry. The amount of data in the pharmaceutical industry is extremely large. If this data is used to train artificial intelligence algorithm models and then used in the process of pharmaceutical research and development, it can effectively speed up the research and development, reduce research and development costs, and increase the success rate of research and development.

[0004] The existing drug reuse discovery methods can be summarized into two categories: the first category is based on similarity calculation, assuming that similar drugs are suitable for similar diseases without obtaining drug-related network structure information; the second category is based on drug-related multi-domain network structure information, ignoring domain-specific features, projecting feature data from different domains into the same vector space, but only processing multiple information into a single whole graph, ignoring the exclusive features of different domains. Obviously, the existing technology lacks the ability to predict new interaction relationships between diseases and drugs in an unsupervised manner. Summary of the invention

[0005] In view of the above problems, the present invention proposes a drug reuse prediction method, device, storage medium and computer equipment. By combining artificial intelligence technology in the process of drug reuse, the concept of multi-part graph is introduced, and the network graph features of proteins, drugs, diseases, metabolites and genes are integrated and fully extracted, and the node features of drugs and diseases are updated, and finally the interaction between drugs and diseases is predicted.

[0006] According to a first aspect of the present invention, a method for predicting drug reuse is provided, comprising:

[0007] Collecting unimodal data and multimodal association relationship data; the unimodal data is feature data used to characterize a class of biological entities, including protein data, drug data, disease data, metabolite data and gene data; the multimodal association relationship data is feature data used to characterize the relationship between different classes of biological entities;

[0008] Using GCN graph neural network technology to construct a bimodal sub-graph based on the unimodal data and the multimodal association relationship data; the bimodal sub-graph is used to characterize the association relationship between the two types of unimodal data;

[0009] Based on the bimodal sub-graph, the feature data of all unimodal data is updated using the GAT graph attention network technology;

[0010] The updated drug data and disease data are reconstructed to obtain target drug-disease association data for characterizing the treatment of the target disease by the target drug.

[0011] Optionally, the multimodal association relationship data includes at least one of drug-disease association relationship data, drug-protein association relationship data, disease-protein association relationship data, gene-protein association relationship data, metabolite-gene association relationship data, metabolite-drug association relationship data, and metabolite-disease association relationship data.

[0012] Optionally, the using of the GCN graph neural network technology to construct a bimodal sub-graph based on the unimodal data and the multimodal association relationship data includes:

[0013] Using GCN graph neural network technology respectively, a plurality of unimodal sub-graphs are constructed based on the plurality of unimodal data; wherein the unimodal sub-graphs are composed of a first node and a first connection relationship between the first nodes, the first node is used to represent feature data in the unimodal data, and the first connection relationship is used to represent the association relationship between the feature data in the unimodal data; the unimodal sub-graphs include at least one of a protein sub-graph, a drug sub-graph, a disease sub-graph, a metabolite sub-graph, and a gene sub-graph;

[0014] Using GCN graph neural network technology to aggregate the corresponding two types of single-modal sub-graphs based on the multi-modal association relationship data to generate a bi-modal sub-graph;

[0015] Among them, the bimodal sub-graph is composed of a second node and a second connection relationship between the second node, the second node is used to represent the feature data in two types of single-modal data, and the second connection relationship is used to represent the association relationship between the feature data in two types of single-modal data; the bimodal sub-graph includes at least one of a drug-disease sub-graph, a drug-protein sub-graph, a disease-protein sub-graph, a gene-protein sub-graph, a metabolite-gene sub-graph, a metabolite-drug sub-graph, and a metabolite-disease sub-graph.

[0016] Optionally, the updating of feature data for all unimodal data using GAT graph attention network technology based on the bimodal sub-graph includes:

[0017] Based on the bimodal sub-graph, the GAT graph attention network technology is used to perform vector feature transfer on the second unimodal data feature according to the first unimodal data feature in the bimodal sub-graph, and the second unimodal data feature is updated; wherein the first unimodal data feature is the feature data of the unimodal data corresponding to a type of biological entity in the bimodal sub-graph; and the second unimodal data feature is the feature data of the unimodal data corresponding to another type of biological entity in the bimodal sub-graph;

[0018] The updated second unimodal data feature is used to perform multiple rounds of iterative updates on other bimodal sub-graphs related to the second unimodal data feature, thereby completing the feature data update for all unimodal data.

[0019] Optionally, the performing vector feature transfer on the second unimodal data feature according to the first unimodal data feature in the bimodal sub-graph by using the GAT graph attention network technology based on the bimodal sub-graph to update the second unimodal data feature includes:

[0020] Using a graph attention mechanism, an aggregation operation of adaptively assigning weights to neighboring second nodes corresponding to the second unimodal data feature through the second node of the first unimodal data feature;

[0021] The second unimodal data feature node set is updated using two bimodal sub-graphs corresponding to the second unimodal data features; wherein the second unimodal data feature node set includes at least one of a drug data feature node set, a protein data feature node set, a disease data feature node set, a metabolite data feature node set, and a gene data feature node set.

[0022] Optionally, the using the updated second unimodal data feature to perform multiple rounds of iterative updates on other bimodal sub-graphs related to the second unimodal data feature to complete the feature data update for all unimodal data includes:

[0023] Using the drug data feature node set and the disease data feature node set, the protein data feature node set and the metabolite data feature node set are updated;

[0024] Using the updated protein data feature node set and metabolite data feature node set to update the gene data feature node set;

[0025] Using the updated gene data feature node set, the protein data feature node set and the metabolite data feature node set are updated;

[0026] The updated protein data feature node set and metabolite data feature node set are used to update the drug data feature node set and the disease data feature node set respectively.

[0027] Optionally, the reconstructing the association relationship between the updated drug data and disease data to obtain target drug-disease association relationship data for characterizing the treatment of the target drug for the target disease includes:

[0028] Using MLP to apply a multi-layer perceptron to construct a drug reuse prediction model, and using cross entropy loss as a loss function to optimize the parameters of the drug reuse prediction model;

[0029] The optimized drug reuse prediction model is used to reconstruct the association relationship between the updated drug data and disease data to obtain target drug-disease association relationship data for characterizing the treatment of the target drug for the target disease.

[0030] According to a second aspect of the present invention, a drug reuse prediction device is provided, comprising:

[0031] A data collection module, used to collect unimodal data and multimodal association relationship data; the unimodal data is feature data used to characterize a type of biological entity, including protein data, drug data, disease data, metabolite data and gene data; the multimodal association relationship data is feature data used to characterize the relationship between different types of biological entities;

[0032] A sub-graph construction module, used for constructing a bimodal sub-graph based on the unimodal data and the multimodal association relationship data using GCN graph neural network technology; the bimodal sub-graph is used to characterize the association relationship between two types of unimodal data;

[0033] A multi-part graph updating module, used to complete feature data update for all unimodal data based on the bimodal part graph using GAT graph attention network technology;

[0034] The drug reuse prediction module is used to reconstruct the association relationship between the updated drug data and disease data to obtain target drug-disease association data for characterizing the treatment of the target drug for the target disease.

[0035] According to a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the drug reuse prediction method as described in any one of the first aspect of the present invention are implemented.

[0036] According to the fourth aspect of the present invention, a computer device is proposed, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the drug reuse prediction method as described in any one of the first aspect of the present invention are implemented.

[0037] The present invention provides a drug reuse prediction method, device, storage medium and computer equipment, which collects unimodal data and multimodal association relationship data, uses GCN graph neural network technology to construct a bimodal sub-graph based on the unimodal data and the multimodal association relationship data, and uses GAT graph attention network technology to complete the feature data update for all unimodal data based on the bimodal sub-graph, and reconstructs the association relationship between the updated drug data features and disease data features to obtain target drug-disease association relationship data for characterizing the treatment of the target drug for the target disease. The present invention improves the intelligence and accuracy of the prediction of the relationship between drugs and diseases by integrating the features of multimodal data and combining multiple graph features.

[0038] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below.

[0039] Based on the following detailed description of specific embodiments of the present invention in conjunction with the accompanying drawings, those skilled in the art will become more aware of the above and other objects, advantages and features of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0041] Figure 1 A schematic diagram of a process flow of a drug reuse prediction method provided by an embodiment of the present invention is shown;

[0042] Figure 2 A schematic diagram of a GCN graph convolutional neural network architecture provided by an embodiment of the present invention is shown;

[0043] Figure 3 A schematic diagram of a process for updating a single-modal data feature according to an embodiment of the present invention is shown;

[0044] Figure 4 A schematic diagram showing the structure of a drug reuse prediction device provided by an embodiment of the present invention is shown;

[0045] Figure 5 A schematic diagram showing the structure of a drug reuse prediction device provided by another embodiment of the present invention is shown;

[0046] Figure 6A schematic diagram of the physical structure of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0047] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0048] The embodiment of the present invention provides a method for predicting drug reuse. Figure 1 As shown, the method may at least include the following steps S101-S104:

[0049] Step S101, collecting unimodal data and multimodal association relationship data.

[0050] Among them, unimodal data is feature data that characterizes information of a class of biological entities, including at least one of protein data, drug data, disease data, metabolite data, and gene data; multimodal association data is feature data that characterizes the relationship between different types of biological entities, including at least one of drug-disease association data, drug-protein association data, disease-protein association data, gene-protein association data, metabolite-gene association data, metabolite-drug association data, and metabolite-disease association data. Generally speaking, data features can be collected by accessing existing medical databases, for example, drug-disease databases such as repoDB, drug-protein databases DGIdb, disease-protein databases DisGeNet, protein-protein databases String, and so on.

[0051] Step S102: Use GCN graph neural network technology to construct a bimodal sub-graph based on unimodal data and multimodal association relationship data.

[0052] Among them, the bimodal sub-graph is used to characterize the association relationship between two types of unimodal data. GCN graph convolutional neural network technology can be used as a feature extractor for graph data, using the feature data in unimodal data and multimodal association relationship data to complete node classification, edge prediction and other functions to achieve the construction of bimodal sub-graph.

[0053] like Figure 2The figure shows the GCN graph convolutional neural network architecture, which follows the message passing framework and consists of two parts: the message passing stage and the output stage. The message passing stage is to update the feature vector of each node by aggregating the neighborhood features of the node, and the readout stage is to generate the features of the entire graph level through a global pooling module. The graph convolutional neural network uses the message passing function to iteratively run the graph convolution operation, so that the feature information can be propagated over a long distance, so that neighborhood features of different ranges can be learned. In an embodiment of the present invention, unimodal data or multimodal association relationship data can be used as graph data, and the feature data contained therein can be used as N nodes (node), each node has its own features, and these node features can form an N×DN×D-dimensional matrix xx, and the relationship between each node can also form an N×NN×N-dimensional matrix aa, called an adjacency matrix. xx and aa are the input X of the GCN graph convolutional neural network. Then, multiple graph convolution operations are performed using a nonlinear activation function through the convolution layer to extract useful node or edge features to complete node classification or link prediction based on nodes and edges, and finally obtain the output result Y. In the neural network layer of the GCN graph convolutional neural network, feature propagation between layers is performed using the following formula:

[0054]

[0055] in: =A+I, I is the identity matrix; yes The degree matrix of ; H is the feature of each layer. For the input layer, H is X; σ is the nonlinear activation function.

[0056] When constructing sub-graphs, the unimodal data and multimodal association relationship data can be processed through the GCN graph convolutional neural network technology to generate unimodal sub-graphs for representing the association relationships of the same type of biological entities and bimodal sub-graphs for representing the association relationships of different types of entities.

[0057] Step S103, based on the bimodal sub-graph, the feature data of all unimodal data is updated using the GAT graph attention network technology.

[0058] After completing the construction of the bimodal sub-graph, the GAT graph attention network technology is used to transfer the vector features of the second unimodal data features according to the first unimodal data features in the bimodal sub-graph to update the second unimodal data features.

[0059] GAT graph attention network technology is a kind of graph attention mechanism (AttentionMechanism) in the process of message transmission. The method of selectively aggregating neighbor node information is based on the principle of using the core graph attention layer to assign different weights to each node in the graph according to the characteristics of the neighboring nodes. Assuming a graph has N nodes, the F-dimensional feature set of the node can be expressed as:

[0060] ,

[0061] The attention layer performs self-attention processing to output an updated set of node features:

[0062] , .

[0063] During the calculation process, self-attention is used to assign attention weights to each node. The calculation formula is:

[0064]

[0065] in × → The mapping of is a weight matrix.

[0066] Finally, self-attention processing is normalized:

[0067]

[0068] in For a feed-forward neural network Parameters, is the activation function. Now we can get :

[0069]

[0070] The bimodal subgraph is used to characterize the association relationship between different types of entities. For example, the drug-protein subgraph is used to characterize the association relationship between drug data and protein data. The drug data feature node set or the protein data feature node set in the drug-protein subgraph can be used as the first single-modal data feature or the second single-modal data feature. The GAT graph attention network technology is used to update the drug data according to the protein data, that is, the neighbor nodes are aggregated through the attention mechanism to achieve adaptive allocation of weights of different neighbor nodes, and the vector feature transfer based on protein data for drug data is completed to update the drug data. Similarly, this method can be used to update the feature data of other categories of modal data based on the updated drug data, and multiple rounds of iterative updates can be repeated until the feature data of all modal data are updated.

[0071] Step S104, reconstructing the association relationship between the updated drug data and disease data to obtain target drug-disease association relationship data for characterizing the treatment of the target disease by the target drug.

[0072] The updated drug data and disease data are rebuilt into associations to obtain target drug-disease association data, which are then used to predict new connections between diseases and drugs to achieve drug reuse.

[0073] The drug reuse prediction method provided by the embodiment of the present invention collects unimodal data and multimodal association relationship data, uses GCN graph neural network technology to construct a bimodal sub-graph based on the unimodal data and multimodal association relationship data, and uses GAT graph attention network technology based on the bimodal sub-graph to complete the feature data update for all unimodal data, and reconstructs the association relationship of the updated drug data and disease data to obtain the target drug-disease association relationship data for characterizing the treatment of the target disease by the target drug. The present invention integrates the features of multimodal data, combines multiple graph features, and introduces the GAT graph attention network mechanism to associate the node relationship between two types of biological entities, thereby overcoming the characteristics that the traditional GCN can only process biological entities with the same attributes and is difficult to describe the node relationship between two independent biological entities, and greatly improves the intelligence and accuracy of the prediction of the relationship between drugs and diseases.

[0074] Optionally, a bimodal sub-graph is constructed based on the unimodal data and the multimodal association relationship data using GCN graph neural network technology, including:

[0075] Using GCN graph neural network technology respectively to construct multiple unimodal sub-graphs based on multiple unimodal data; wherein the unimodal sub-graphs are composed of first nodes and first connection relationships between the first nodes, the first nodes are used to represent feature data in the unimodal data, and the first connection relationships are used to represent the association relationship between the feature data in the unimodal data; the unimodal sub-graphs may include protein sub-graphs, drug sub-graphs, disease sub-graphs, metabolite sub-graphs, gene sub-graphs, etc.;

[0076] The GCN graph neural network technology is used to aggregate the corresponding two types of unimodal sub-graphs based on the multimodal association relationship data to generate a bimodal sub-graph; wherein the bimodal sub-graph is composed of a second node and a second connection relationship between the second node, the second node is used to represent the feature data in the two types of unimodal data, and the second connection relationship is used to represent the association relationship between the feature data in the two types of unimodal data; the bimodal sub-graph may include a drug-disease sub-graph, a drug-protein sub-graph, a disease-protein sub-graph, a gene-protein sub-graph, a metabolite-gene sub-graph, a metabolite-drug sub-graph, a metabolite-disease sub-graph, and the like.

[0077] For a unimodal subgraph, feature nodes and edges between feature nodes may be included, and the edges are used to establish the connection relationship between the feature nodes. ∈ It can be expressed as ( ), consisting of a node set and an edge set composition, ∈ Is a node Each unimodal graph describes a class of biological entities. They represent protein sub-map, drug sub-map, disease sub-map, metabolite sub-map, and gene sub-map respectively. Represents the feature node set of protein data, Represents the drug data feature node set, Represents the disease data feature node set, Represents the metabolite data feature node set, Represents a set of characteristic nodes of gene data. Similarly, a unimodal subgraph used to describe multiple types of biological entities can constitute a multimodal subgraph. The bimodal subgraph in the embodiment of the present invention is composed of two types of unimodal data. On the basis of the two unimodal subgraphs, multimodal association relationship data is introduced to generate a bimodal subgraph. The edges in the bimodal subgraph only exist between different types of modal data. In other words, the bimodal subgraph is used to characterize the association relationship between different types of modal data. For example, in the constructed protein subgraph And drug sub-map Based on this, drug-protein association data is introduced to generate a bimodal sub-graph ,in Represents a set of nodes for two types of modal data, such as protein and disease. All edges of the bimodal graph exist only in , express The edge between ∈ Is a node The characteristic data of the unimodal data, ∈ Is a node The feature data of the unimodal data.

[0078] Optionally, step S103: based on the bimodal sub-graph, the feature data update for all unimodal data is completed by using the GAT graph attention network technology, which may specifically include steps S103-1 to S103-2:

[0079] Step S103-1: Based on the bimodal sub-graph, the GAT graph attention network technology is used to perform vector feature transfer on the second unimodal data feature according to the first unimodal data feature in the bimodal sub-graph, and the second unimodal data feature is updated; wherein the first unimodal data feature is feature data of unimodal data corresponding to a type of biological entity in the bimodal sub-graph; and the second unimodal data feature is feature data of unimodal data corresponding to another type of biological entity in the bimodal sub-graph;

[0080] Step S103 - 2 : performing multiple rounds of iterative updates on other bimodal sub-graphs related to the second unimodal data features using the updated second unimodal data features, and completing feature updates for all unimodal data features.

[0081] Among them, step S103-1: based on the bimodal sub-graph, using the GAT graph attention network technology to perform vector feature transfer on the second unimodal data feature according to the first unimodal data feature in the bimodal sub-graph, and updating the second unimodal data feature, may specifically include the following methods:

[0082] The graph attention mechanism is used to perform an aggregation operation of adaptive weight allocation on the neighbor second nodes corresponding to the second unimodal data feature through the second node of the first unimodal data feature. Specifically, the following formula can be used:

[0083]

[0084]

[0085]

[0086] in, represents the weight of the second node of the neighbor; bg represents bipartite graph convolution; agg represents aggregation; represents the activation function; Represented in the bimodal sub-graph ) in the second node Neighbor; represents a learnable matrix; || represents the connection of two matrices; Represents the linear rectification function.

[0087] At the same time, the second unimodal data feature node set is updated using two bimodal sub-graphs corresponding to the second unimodal data feature. Among them, the two bimodal sub-graphs corresponding to the second unimodal data feature refer to two different bimodal sub-graphs containing the second unimodal data feature. The second unimodal data feature node set may include a drug data feature node set, a protein data feature node set, a disease data feature node set, a metabolite data feature node set, and a gene data feature node set. For example, taking the example of updating the protein data feature node set using the drug data feature node set and the disease data feature node set based on the drug-protein sub-graph and the disease-protein sub-graph, the following formula can be used specifically:

[0088]

[0089]

[0090] in, represent Gather to Vector feature transfer of the collection.

[0091] Optionally, step S103-2: using the updated second unimodal data features to perform multiple rounds of iterative updates on other bimodal sub-graphs related to the second unimodal data features, to complete the feature data update for all unimodal data, such as Figure 3 As shown, the following steps can be used:

[0092] Step S103-2-1: using the drug data feature node set and the disease data feature node set to update the protein data feature node set and the metabolite data feature node set;

[0093] Step S103-2-2: using the updated protein data feature node set and metabolite data feature node set to update the gene data feature node set;

[0094] Step S103-2-3: using the updated gene data feature node set to update the protein data feature node set and the metabolite data feature node set;

[0095] Step S103-2-4: respectively update the drug data feature node set and the disease data feature node set using the updated protein data feature node set and metabolite data feature node set.

[0096] It can be understood that the above-mentioned multiple rounds of iterative update steps are the specific update process adopted by the embodiment of the present invention in practical applications. In practical applications, the feature update order of all single-modal data features can be set differently according to actual conditions, and the present invention will not go into details.

[0097] Optionally, the updated drug data and disease data are reconstructed to obtain target drug-disease association data for characterizing the treatment of the target drug for the target disease, which may specifically include:

[0098] A drug reuse prediction model was constructed using MLP and a multi-layer perceptron. The cross-entropy loss was used as the loss function to optimize the parameters of the drug reuse prediction model. The optimized drug reuse prediction model was used to reconstruct the association between the updated drug data and disease data to obtain the target drug-disease association data used to characterize the treatment of the target drug for the target disease.

[0099] In the embodiment of the present invention, the process of reconstructing the drug-disease interaction graph by using the MLP application multilayer perceptron is to use the encoder in the MLP application multilayer perceptron to return the embedding vectors of the drug data feature node set and the disease data feature node set, connect the two embedding vectors to represent the drug data feature-disease data feature pair, input the decoder, and the decoder scores the interaction relationship between the drug data feature-disease data feature pair through the three-layer MLP neural network, that is, calculate the sigmoid function The probability of an edge between The probability of the edge between is used to characterize The size of the interaction between the drug data feature and the disease data feature pair indicates the probability that the drug can treat the disease. The target drug-disease association data in the embodiment of the present invention is used to characterize the treatment of the target drug for the target disease, which can be expressed by the treatment probability. The specific formula is as follows:

[0100]

[0101]

[0102] in, express The probability of the edge between . When training the model, the cross entropy loss is used as the loss function to optimize the model parameters. The cross entropy loss is to randomly sample an edge and calculate the difference between it and the true edge, that is, to calculate the final loss function. By minimizing the loss function value, the prediction results will become more and more accurate as the training progresses.

[0103] The drug reuse prediction method proposed in the embodiment of the present invention collects unimodal data and multimodal association relationship data, uses GCN graph neural network technology to construct a bimodal sub-graph based on the unimodal data and multimodal association relationship data, and uses GAT graph attention network technology based on the bimodal sub-graph to complete the feature data update for all unimodal data, and reconstructs the association relationship of the updated drug data and disease data to obtain the target drug-disease association relationship data used to characterize the treatment of the target drug for the target disease. The present invention integrates the characteristics of multimodal data, combines a variety of network structure characteristics and exclusive characteristics in the field of biological categories, and introduces the GAT graph attention network mechanism to associate the node relationship between two types of biological entities, overcomes the characteristics of traditional GCN that can only process biological entities with the same attributes and is difficult to describe the node relationship between two independent biological entities, and greatly improves the intelligence and accuracy of the prediction of the relationship between drugs and diseases.

[0104] Further, as Figure 1 The specific implementation of the present invention provides a drug reuse prediction device, such as Figure 4 As shown, the apparatus may include: a data collection module 410 , a sub-graph construction module 420 , a multi-sub-graph updating module 430 and a drug reuse prediction module 440 .

[0105] The data collection module 410 can be used to collect unimodal data and multimodal association relationship data; unimodal data is feature data used to characterize a type of biological entity, including protein data, drug data, disease data, metabolite data, and gene data; multimodal association relationship data is feature data used to characterize the relationship between different types of biological entities;

[0106] The sub-graph construction module 420 can be used to construct a bimodal sub-graph based on the unimodal data and the multimodal association relationship data using the GCN graph neural network technology; the bimodal sub-graph is used to represent the association relationship between two types of unimodal data;

[0107] The multi-part graph updating module 430 can be used to complete the feature data update for all unimodal data based on the bimodal part graph using the GAT graph attention network technology;

[0108] The drug reuse prediction module 440 can be used to reconstruct the association relationship between the updated drug data and disease data to obtain target drug-disease association relationship data for characterizing the treatment of the target drug for the target disease.

[0109] Optionally, an embodiment of the present invention provides a drug reuse prediction device, such as Figure 5 As shown, the sub-graph construction module 420 may further include a single-sub-graph construction unit 421 and a bipartite graph construction unit 422:

[0110] The single-sub-graph construction unit 421 can be used to construct multiple single-modal sub-graphs based on multiple single-modal data using GCN graph neural network technology respectively; wherein the single-modal sub-graph is composed of a first node and a first connection relationship between the first nodes, the first node is used to represent the feature data of the single-modal data, and the first connection relationship is used to represent the association relationship between the feature data in the single-modal data; the single-modal sub-graph includes at least one of a protein sub-graph, a drug sub-graph, a disease sub-graph, a metabolite sub-graph, and a gene sub-graph.

[0111] The bipartite graph construction unit 422 can be used to use the GCN graph neural network technology to aggregate the corresponding two types of unimodal sub-graphs based on the multimodal association relationship data to generate a bimodal sub-graph; wherein the bimodal sub-graph is composed of a second node and a second connection relationship between the second node, the second node is used to represent the feature data of the two types of unimodal data, and the second connection relationship is used to represent the association relationship between the feature data in the two types of unimodal data; the bimodal sub-graph includes at least one of a drug-disease sub-graph, a drug-protein sub-graph, a disease-protein sub-graph, a gene-protein sub-graph, a metabolite-gene sub-graph, a metabolite-drug sub-graph, and a metabolite-disease sub-graph.

[0112] Optionally, the multi-part graph updating module 430 may include a vector feature transfer unit 431 and a feature iterative updating unit 432:

[0113] The vector feature transfer unit 431 can be used to perform vector feature transfer on the second unimodal data feature according to the first unimodal data feature in the bimodal sub-graph using the GAT graph attention network technology based on the bimodal sub-graph, and update the second unimodal data feature; wherein the first unimodal data feature is the feature data of the unimodal data corresponding to a type of biological entity in the bimodal sub-graph; and the second unimodal data feature is the feature data of the unimodal data corresponding to another type of biological entity in the bimodal sub-graph;

[0114] The feature iterative updating unit 432 is used to perform multiple rounds of iterative updates on other bimodal sub-graphs related to the second unimodal data features using the updated second unimodal data features, so as to complete the feature data update for all unimodal data.

[0115] Optionally, the vector feature transfer unit 431 can also be used to utilize the graph attention mechanism to perform an aggregation operation of adaptive weight allocation on the neighbor second nodes corresponding to the second unimodal data feature through the second node of the first unimodal data feature; and to update the second unimodal data feature node set using two bimodal sub-graphs corresponding to the second unimodal data feature; wherein the second unimodal data feature node set includes at least one of a drug data feature node set, a protein data feature node set, a disease data feature node set, a metabolite data feature node set, and a gene data feature node set.

[0116] Optionally, the feature iteration updating unit 432 can also be used to update the protein data feature node set and the metabolite data feature node set using the drug data feature node set and the disease data feature node set; update the gene data feature node set using the updated protein data feature node set and the metabolite data feature node set; update the protein data feature node set and the metabolite data feature node set using the updated gene data feature node set; and update the drug data feature node set and the disease data feature node set using the updated protein data feature node set and the metabolite data feature node set, respectively.

[0117] Optionally, the drug reuse prediction module 440 may further include a model building unit 441 and a model prediction unit 442:

[0118] The model building unit 441 can be used to build a drug reuse prediction model using an MLP application multi-layer perceptron, and use the cross entropy loss as a loss function to optimize the parameters of the drug reuse prediction model.

[0119] The model prediction unit 442 can be used to use the drug reuse prediction model to reconstruct the association relationship between the updated drug data and disease data, and obtain target drug-disease association data for characterizing the treatment of the target drug for the target disease.

[0120] It should be noted that for other corresponding descriptions of the functional modules involved in the drug reuse prediction device provided in the embodiment of the present invention, reference can be made to Figure 1 The corresponding description of the method shown will not be repeated here.

[0121] Based on the above Figure 1 The method shown, accordingly, an embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the drug reuse prediction method described in any of the above embodiments are implemented.

[0122] Based on the above Figure 1 The method shown and Figure 5 The embodiment of the system shown in the figure, the embodiment of the present invention also provides a physical structure diagram of a computer device, such as Figure 6 As shown, the computer device may include a communication bus, a processor, a memory and a communication interface, and may also include an input / output interface and a display device, wherein each functional unit may communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the steps of the drug reuse prediction method described in the above embodiment.

[0123] Those skilled in the art can clearly understand that the specific working processes of the systems, devices, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and for the sake of brevity, they are not further described here.

[0124] In addition, the functional units in various embodiments of the present invention may be physically independent of each other, or two or more functional units may be integrated together, or all functional units may be integrated into one processing unit. The above integrated functional units may be implemented in the form of hardware, or in the form of software or firmware.

[0125] A person skilled in the art can understand that if the integrated functional unit is implemented in the form of software and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, which includes several instructions to enable a computing device (such as a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention when running the instructions. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0126] Alternatively, all or part of the steps of implementing the aforementioned method embodiments may be accomplished by hardware associated with program instructions (such as a computing device such as a personal computer, a server, or a network device), and the program instructions may be stored in a computer-readable storage medium. When the program instructions are executed by a processor of the computing device, the computing device executes all or part of the steps of the method described in the various embodiments of the present invention.

[0127] In addition, the genetic data used in the embodiments of this application are obtained under the premise of strictly following the informed consent of the user. Ensure that the source of the data is legal, the collection process is standardized, and its use is in full compliance with the relevant national and local laws and regulations on genetic data protection and management, aiming to protect the privacy rights of users while promoting the healthy development of scientific research and medical fields.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that within the spirit and principles of the present invention, the technical solutions described in the aforementioned embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate from the protection scope of the present invention.

Claims

1. A drug reuse prediction method, characterized in that: include: Collecting unimodal data and multimodal association data; the unimodal data is feature data used to characterize a class of biological entities, including protein data, drug data, disease data, metabolite data and gene data; the multimodal association data is feature data used to characterize the relationship between different types of biological entities, and the multimodal association data includes at least one of drug-disease association data, drug-protein association data, disease-protein association data, gene-protein association data, metabolite-gene association data, metabolite-drug association data and metabolite-disease association data; A bimodal subgraph is constructed based on the unimodal data and the multimodal association relationship data using GCN graph neural network technology; the bimodal subgraph is used to characterize the association relationship between the two types of unimodal data; wherein the bimodal subgraph is composed of a second node and a second connection relationship between the second node, the second node is used to represent the feature data in the two types of unimodal data, and the second connection relationship is used to represent the association relationship between the feature data in the two types of unimodal data; the bimodal subgraph includes at least one of a drug-disease subgraph, a drug-protein subgraph, a disease-protein subgraph, a gene-protein subgraph, a metabolite-gene subgraph, a metabolite-drug subgraph, and a metabolite-disease subgraph; Based on the bimodal sub-graph, the GAT graph attention network technology is used to complete the feature data update for all unimodal data, including: based on the bimodal sub-graph, the GAT graph attention network technology is used to perform vector feature transfer on the second unimodal data feature according to the first unimodal data feature in the bimodal sub-graph, and the second unimodal data feature is updated; wherein, the first unimodal data feature is the feature data of the unimodal data corresponding to a type of biological entity in the bimodal sub-graph; the second unimodal data feature is the feature data of the unimodal data corresponding to another type of biological entity in the bimodal sub-graph; and the updated second unimodal data feature is used to perform multiple rounds of iterative updates on other bimodal sub-graphs related to the second unimodal data feature, and the feature data update for all unimodal data is completed; The updated drug data and disease data are reconstructed to obtain target drug-disease association data for characterizing the treatment of the target disease by the target drug.

2. The method according to claim 1, characterized in that The method of using the GCN graph neural network technology to construct a bimodal sub-graph based on the unimodal data and the multimodal association relationship data includes: Using GCN graph neural network technology respectively, multiple unimodal sub-graphs are constructed based on multiple unimodal data; wherein the unimodal sub-graphs are composed of first nodes and first connection relationships between the first nodes, the first nodes are used to represent feature data in the unimodal data, and the first connection relationships are used to represent the association relationship between the feature data in the unimodal data; the unimodal sub-graphs include at least one of a protein sub-graph, a drug sub-graph, a disease sub-graph, a metabolite sub-graph, and a gene sub-graph; The GCN graph neural network technology is used to aggregate the corresponding two types of unimodal sub-graphs based on the multimodal association relationship data to generate a bimodal sub-graph.

3. The method according to claim 1, characterized in that The method of performing vector feature transfer on the second unimodal data feature according to the first unimodal data feature in the bimodal sub-graph by using the GAT graph attention network technology based on the bimodal sub-graph to update the second unimodal data feature includes: Using a graph attention mechanism, an aggregation operation of adaptively assigning weights to neighboring second nodes corresponding to the second unimodal data feature through the second node of the first unimodal data feature; The second unimodal data feature node set is updated using two bimodal sub-graphs corresponding to the second unimodal data features; wherein the second unimodal data feature node set includes at least one of a drug data feature node set, a protein data feature node set, a disease data feature node set, a metabolite data feature node set, and a gene data feature node set.

4. The method according to claim 3, characterized in that The method of performing multiple rounds of iterative updates on other bimodal sub-graphs related to the second bimodal data features by using the updated second bimodal data features to complete the feature data update for all bimodal data includes: Using the drug data feature node set and the disease data feature node set, the protein data feature node set and the metabolite data feature node set are updated; Using the updated protein data feature node set and metabolite data feature node set to update the gene data feature node set; Using the updated gene data feature node set, the protein data feature node set and the metabolite data feature node set are updated; The updated protein data feature node set and metabolite data feature node set are used to update the drug data feature node set and the disease data feature node set respectively.

5. The method according to any one of claims 1 to 4, characterized in that: The reconstructing of the association relationship between the updated drug data and disease data to obtain target drug-disease association relationship data for characterizing the treatment of the target drug for the target disease includes: Using MLP to apply a multi-layer perceptron to construct a drug reuse prediction model, and using cross entropy loss as a loss function to optimize the parameters of the drug reuse prediction model; The optimized drug reuse prediction model is used to reconstruct the association relationship between the updated drug data and disease data to obtain target drug-disease association relationship data for characterizing the treatment of the target drug for the target disease.

6. A drug reuse prediction device, characterized in that: include: A data collection module, used to collect unimodal data and multimodal association relationship data; the unimodal data is feature data used to characterize a class of biological entities, including protein data, drug data, disease data, metabolite data and gene data; the multimodal association relationship data is feature data used to characterize the relationship between different types of biological entities, and the multimodal association relationship data includes at least one of drug-disease association relationship data, drug-protein association relationship data, disease-protein association relationship data, gene-protein association relationship data, metabolite-gene association relationship data, metabolite-drug association relationship data, and metabolite-disease association relationship data; A sub-graph construction module, used for constructing a bimodal sub-graph based on the unimodal data and the multimodal association relationship data using GCN graph neural network technology; the bimodal sub-graph is used to characterize the association relationship between two types of unimodal data; wherein the bimodal sub-graph is composed of a second node and a second connection relationship between the second node, the second node is used to represent feature data in the two types of unimodal data, and the second connection relationship is used to represent the association relationship between the feature data in the two types of unimodal data; the bimodal sub-graph includes at least one of a drug-disease sub-graph, a drug-protein sub-graph, a disease-protein sub-graph, a gene-protein sub-graph, a metabolite-gene sub-graph, a metabolite-drug sub-graph, and a metabolite-disease sub-graph; A multi-part graph updating module, used to complete feature updates for all unimodal data based on the bimodal part graphs using the GAT graph attention network technology; A drug reuse prediction module is used to reconstruct the association relationship between the updated drug data and disease data to obtain target drug-disease association relationship data for characterizing the treatment of the target drug for the target disease; The multi-part graph updating module includes a vector feature transfer unit and a feature iterative updating unit; The vector feature transfer unit is used to perform vector feature transfer on the second unimodal data feature according to the first unimodal data feature in the bimodal sub-graph using the GAT graph attention network technology based on the bimodal sub-graph, and update the second unimodal data feature; wherein the first unimodal data feature is feature data of unimodal data corresponding to a type of biological entity in the bimodal sub-graph; and the second unimodal data feature is feature data of unimodal data corresponding to another type of biological entity in the bimodal sub-graph; The feature iterative updating unit is used to perform multiple rounds of iterative updating on other bimodal sub-graphs related to the second bimodal data features using the updated second bimodal data features, so as to complete the feature data update for all bimodal data.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the drug reuse prediction method according to any one of claims 1 to 5 are implemented.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the step of drug reuse prediction according to any one of claims 1 to 5 is implemented.

Images