Drug relocation method and system based on subgraph perception and mixed graph neural network

Abstract

The invention discloses a drug relocation method and system based on subgraph perception and a mixed graph neural network, and the method comprises the steps: generating a corresponding dynamic drug-disease associated subgraph # imgabs1 # for a target drug-disease pair in a drug-disease associated network # imgabs0 #; aiming at the medicine nodes and the disease nodes in the # imgabs2, respectively constructing a medicine sub-graph # imgabs3 # and a disease sub-graph # imgabs4 #; performing deep feature extraction on # imgabs5 by applying a multi-head dynamic graph attention neural network model, and performing feature extraction on # imgabs6 and # imgabs7 on the basis of fusion of a GCN model and a GAT model to obtain final feature representations of # imgabs8, # imgabs9 and # imgabs10; and fusing the three, inputting the fused three into a drug-disease association prediction model based on a multi-layer perceptron (MLP) network, and predicting the association relationship of the target drug-disease pair. According to the invention, efficient and accurate drug-disease association prediction can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of drug prediction and analysis, and specifically relates to a drug repositioning method and system based on subgraph perception and hybrid graph neural network. Background Art

[0002] Drug repositioning refers to the process of accelerating drug development by exploring new indications for approved drugs. Compared with traditional new drug research and development, drug repositioning re-studies and analyzes existing drugs to find their potential therapeutic uses beyond their original indications. Since the safety and pharmacokinetic properties of drugs have been verified by previous clinical trials and the market, drug repositioning can not only significantly shorten the development cycle from laboratory research to clinical application, but also effectively reduce research and development costs and the risk of failure, thus becoming an efficient and low-risk alternative in the field of drug research and development. This strategy has shown great potential in therapeutic research in the fields of cancer, neurodegenerative diseases, infectious diseases, etc. Its high success rate and low development cost have brought important commercial value and social benefits to the pharmaceutical industry and medical researchers.

[0003] In recent years, with the rapid development of deep learning technology, especially graph neural networks, graph learning-based drug repositioning methods have gradually become a research hotspot. This type of method can efficiently model the complex relationship between drugs and diseases by utilizing the powerful graph structure data processing capabilities of graph neural networks. For example, graph convolutional networks gradually aggregate node neighborhood information by stacking convolutional layers to achieve multi-level feature expression of drug-disease networks; graph attention networks effectively capture the differences in local structures in the network by adaptively assigning importance weights to different nodes.

[0004] However, the existing drug repositioning methods fail to fully explore the potential information in the drug-disease network and cannot effectively combine the global and local characteristics of the network, which limits the accuracy and reliability of drug-disease relationship prediction. Therefore, how to develop a drug repositioning method that can dynamically capture the local structural characteristics of the network and make full use of global semantic information to achieve efficient and accurate drug-disease association prediction has become an important problem that needs to be solved in the current technical field. Summary of the invention

[0005] The purpose of the present invention is to provide a drug repositioning method based on subgraph perception and hybrid graph neural network, which can achieve efficient and accurate drug-disease association prediction.

[0006] In order to achieve the above-mentioned purpose of the present invention, the technical solution provided by the present invention is:

[0007] In a first aspect, the present application provides a drug repositioning method based on subgraph perception and hybrid graph neural network, comprising:

[0008] S1. In the drug-disease association network In this paper, the target drug-disease pair is the center, from the target drug-disease pair Extract all drug nodes and disease nodes related to the target node in the neighborhood, as well as the edges between these nodes, to generate the corresponding dynamic drug-disease association subgraph ;

[0009] S2, using the similarity data between drug nodes and disease nodes, The drug nodes and disease nodes in the and the disease subgraph ;

[0010] S3, yes The multi-head dynamic graph attention neural network model is used for deep feature extraction to obtain The final feature representation ;

[0011] S4, based on the fusion of the graph convolution network GCN model and the graph attention network GAT model, and The network performs feature extraction and obtains and The final feature representation and ;

[0012] S5. , and The final feature representation , , Fusion to build a unified feature representation ;Will Input a drug-disease association prediction model based on a multi-layer perceptron MLP network, and output the prediction result of the association relationship of the target drug-disease pair through the drug-disease association prediction model;

[0013] The multi-head dynamic graph attention neural network model in step S3, the graph convolutional network GCN model and the graph attention network GAT model in step S4, and the drug-disease association prediction model based on the multi-layer perceptron MLP network in step S5 are obtained through training and optimization based on the known target drug-disease pair association relationship.

[0014] In a possible implementation, the method further includes: based on the known target drug-disease pair association relationship, training and optimizing the multi-head dynamic graph attention neural network model in step S3, the graph convolution network GCN model and the graph attention network GAT model in step S4, and the drug-disease association prediction model based on the multi-layer perceptron MLP network in step S5, including:

[0015] S1 to S5 are performed on the target drug-disease pair with known association to obtain the association prediction result of the target drug-disease pair; according to the association prediction result of the target drug-disease pair and the true label of the association of the target drug-disease pair, the cross entropy loss function is used as the loss function, and a regularization term is introduced to improve the generalization ability of the model. At the same time, the Adam optimizer is combined to update the learnable parameters to accelerate the convergence of the model, and the Dropout technology is used to reduce overfitting to achieve model optimization.

[0016] As a result, when processing sparse data or unbalanced data sets, problems such as overfitting or insufficient generalization ability can be avoided, the prediction effect of rare diseases can be improved, and the robustness to noise information in the data can be improved.

[0017] In a possible implementation, in step S1, the ;

[0018] Constructing a drug-disease association network using known drug-disease associations ,in is the drug node set, Include Different drugs, is the set of disease nodes, Include different diseases; is the edge set, ,in Represents drug node and disease nodes If the relationship between drugs is known and disease If there is a relationship, then the edge The weight of Set to 1, otherwise 0.

[0019] In a possible implementation, in step S2, the drug subgraph ,in for The node collection in contains All drug nodes in for The edge set in , represents Similarity data between drug nodes in the disease subgraph ),in for The node collection in contains All disease nodes in for The edge set in , represents Similarity data between disease nodes in ;

[0020] In step S2, based on the chemical structure information (such as SMILES representation) obtained from the DrugBank database, the similarity between all drug nodes is calculated using the Tanimoto similarity formula to generate a drug similarity matrix. , Middle Line Elements of a column Indicates Drug nodes and Drug similarity matrix Represents the similarity data between drug nodes;

[0021] In step S2, the phenotypic description of the disease (such as MeSH terms) is obtained from the OMIM disease phenotype database, and the similarity between all disease nodes is obtained using a semantic calculation method based on Jaccard similarity. , thus generating a disease similarity matrix , Middle Line Elements of a column Indicates Disease nodes and The similarity between disease nodes; using the disease similarity matrix Represents similarity data between disease nodes;

[0022] Based on the similarity data between drug nodes and the similarity data between disease nodes, the K nearest neighbor similarity matrix of drugs and diseases is constructed respectively. and , Middle Line Elements of a column and Middle Line Elements of a column The definition is as follows:

[0023] ;

[0024] in, Representative Drug nodes Neighbor set, if drug nodes belong to the K nearest neighbor set, the similarity value remains the original similarity value , otherwise set to 0;

[0025] ;

[0026] in, Representative Disease Node Neighbor set, if disease nodes belong to the K nearest neighbor set, the similarity value remains the original similarity value , otherwise set to 0;

[0027] Get the K nearest neighbor similarity matrix of drugs and diseases and After the matrix, the two are respectively used as and The weight of the corresponding edge in .

[0028] In a possible implementation, in step S3, the multi-head dynamic graph attention neural network model includes a multi-layer neural network;

[0029] The step S3 comprises:

[0030] S301, in Layer neural network In an attention head, the node With Node The attention weight is calculated as follows:

[0031] ;

[0032] in, , Representative Node of The set of neighbor nodes in the neighborhood, and Represents nodes and The eigenvector of For the Layer Neural Network The weight matrix of the attention heads is used to map the input features into a high-dimensional latent space. For the Layer neural network The saliency weight vector of the features connected by the attention heads; and The value of can be randomly initialized and changes as the model learns;

[0033] S302. Based on the attention weight, a multi-head attention mechanism is used to perform weighted summation (dynamic aggregation) of the neighboring node features of the target node. The formula is defined as:

[0034] ;

[0035] in, Representation Node In the The updated feature representation in the layer neural network, Represents the number of attention heads;

[0036] In the above steps, each attention head independently calculates the attention weight and averages the features of adjacent nodes to ensure consistent dimensions;

[0037] S303, yes The ReLU activation function is used for nonlinear activation and the normalization function is used for normalization to enhance the nonlinear ability of feature expression and obtain the node In the Output features of layer neural network ;

[0038] S304, the output features of the multi-layer neural network are integrated to obtain The characteristic matrix ; The feature matrix No. Behavior , ;

[0039] That is to say, Each row in corresponds to the feature vector of a node after training;

[0040] S305, using a differentiable pooling method to learn trainable normalized weights, grouping and aggregating node features to reduce the size of the drug-disease subgraph and gradually expand the receptive field range of the node; The characteristic matrix and the adjacency matrix Mapped to smaller subgraphs via the following transformation:

[0041] ;

[0042] ;

[0043] in, is a learnable normalized weight matrix used to distribute the original nodes into fewer super nodes, The dimension is , is the original number of nodes, is the number of super nodes after pooling, where the element Represents the original node Assign to super nodes The weight of Initialized randomly and changes as the model learns; for The adjacency matrix describes the connection relationship between nodes, where the elements are The weight of the corresponding edge in ; Represents the feature matrix of the super node, It is the pooled adjacency matrix, describing the relationship between super nodes;

[0044] and constitute The final feature representation .

[0045] In a possible implementation, step S4 includes:

[0046] S401, using the GCN model to aggregate the features of the target node and its adjacent nodes; wherein the GCN model includes layer neural network, where 2. Drug Node and disease nodes In the The feature update formula of the layer neural network is:

[0047] ;

[0048] ;

[0049] in, , Respectively represent Layer Neural Network Drug Node and disease nodes The local structural characteristics, normalized coefficient and Defined as:

[0050] ;

[0051] ;

[0052] in, and Represents drug nodes and drug nodes In the drug subgraph The degree in (the sum of in-degree and out-degree); and Represents disease nodes and disease nodes In the disease subgraph The degree in ;

[0053] , In the GCN model, Layer Neural Network for Drug Subgraphs and the disease subgraph The learnable parameter matrix, whose values ​​can be randomly initialized and optimized by the back-propagation algorithm during training;

[0054] Calculated by the above formula and ;

[0055] S402, using the GAT model to capture the interaction between adjacent nodes; wherein the GAT model includes layer neural network, where 2; Each layer of the neural network uses a multi-head attention mechanism to focus on the drug nodes and disease nodes Update the characteristics of the drug node and disease nodes In the GAT model The feature update formula of the layer neural network is defined as follows:

[0056] ;

[0057] ;

[0058] and Drug Node and disease nodes In the GAT model The node features after the layer neural network is updated, is the number of attention heads included in the GAT model, In the GAT model The first layer of the neural network Drug nodes in the attention head and drug nodes The attention weights between In the GAT model The first layer of the neural network Disease nodes in attention heads and disease nodes The attention weights between and In the GAT model, Layer Neural Network for Drug Subgraphs and the disease subgraph The learnable parameter matrix, whose values ​​can be randomly initialized and optimized by the back-propagation algorithm during training;

[0059] Calculated by the above formula and ;

[0060] S403, weighted fusion is performed on the outputs of the GCN model and the GAT model to generate the final feature representation of the target node; the fusion process is defined as:

[0061] ;

[0062] ;

[0063] in, and Represents drug nodes and disease nodes The final feature representation is, It is the fusion ratio hyperparameter, which is used to adjust the relative contribution of the output of the GCN model and the GAT model;

[0064] By drug subgraph The final feature representation of all drug nodes in constitutes the drug subgraph The final feature representation ;

[0065] By disease subgraph The final feature representation of all disease nodes in constitutes the disease subgraph The final feature representation .

[0066] In a possible implementation, in step S5, , and The final feature representation , , Fusion to build a unified feature representation , the process of feature fusion is defined as follows:

[0067] ;

[0068] in, Represents a splicing operation.

[0069] In a possible implementation, the The drug-disease association prediction model based on the multi-layer perceptron MLP network is input, and the drug-disease association prediction model outputs the prediction results of the association relationship of the target drug-disease pair, including:

[0070] The output layer of the drug-disease association prediction model predicts the drug node by the following formula and disease nodes The probability of association between them, the greater the probability, the more likely the association is:

[0071] ;

[0072] in, , , and are weight and coefficient parameters, their values ​​can be randomly initialized and they are optimized by the back-propagation algorithm during the training process.

[0073] In one possible implementation, the cross entropy loss function is defined as follows:

[0074] ;

[0075] in, Drug Node and disease nodes The true label of the correlation relationship between Predicted drug nodes and disease nodes The probability of association between is the total number of samples. Among them, the samples are target drug-disease pairs with known associations.

[0076] A regularization term is introduced to improve the generalization ability of the model, including adding a regularization term to the objective function, which is defined as:

[0077] ;

[0078] in, represents the regularization coefficient, Represents the learnable parameters of the k-th layer of the model.

[0079] In the second aspect, the present application provides a drug relocation system based on subgraph perception and hybrid graph neural network, including: a data input layer, a subgraph extraction layer, a feature extraction layer, a multi-head dynamic graph attention neural network model layer, a GCN model and a GAT model fusion neural network layer, and a feature fusion and prediction layer;

[0080] The data input layer is used to collect and integrate data on drugs, diseases and their associations to form a drug-disease association network. ;

[0081] The subgraph extraction layer is used in the drug-disease association network In this paper, the target drug-disease pair is the center, from the target drug-disease pair In the neighborhood, all drug nodes and disease nodes related to the target node, as well as the edges between these nodes, are extracted to generate a dynamic drug-disease association subgraph ;

[0082] The feature extraction layer is used to utilize the similarity data between drug nodes and the similarity data between disease nodes to The drug nodes and disease nodes in the and the disease subgraph ;

[0083] The multi-head dynamic graph attention neural network model layer is used to analyze the dynamic drug-disease association subgraph Perform deep feature extraction to obtain The final feature representation ;

[0084] The GCN model and the GAT model fusion neural network layer are used to fuse the graph convolution network GCN model and the graph attention network GAT model respectively. and The network performs feature extraction and obtains and The final feature representation and ;

[0085] The feature fusion and prediction layer is used to , and The final feature representation , , Fusion to build a unified feature representation ; The drug-disease association prediction model based on the multi-layer perceptron MLP network is input, and the drug-disease association prediction model outputs the prediction result of the association relationship of the target drug-disease pair.

[0086] The system realizes drug relocation based on the above-mentioned drug relocation method based on subgraph perception and hybrid graph neural network.

[0087] In a third aspect, the present application provides an electronic device, including: a memory and a processor;

[0088] The memory is used to store computer programs;

[0089] The processor is used to call the computer program to execute the method as described above.

[0090] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed on an electronic device, the electronic device implements the method described above.

[0091] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is run on an electronic device, the electronic device implements the method as described above.

[0092] The specific implementation methods of the second to fifth aspects of the present application can refer to the implementation methods of the first aspect, which will not be repeated here.

[0093] The beneficial effects of the present invention are as follows:

[0094] The present invention discloses a drug repositioning method based on subgraph perception and hybrid graph neural network, which dynamically extracts subgraphs centered on target drug-disease pairs; constructs drug subgraphs and disease subgraphs using the similarity between drugs and diseases, and comprehensively uses multi-head dynamic graph attention mechanism and multimodal feature extraction technology fused with GCN model and GAT model to effectively capture the local structure and global similarity relationship of nodes; and realizes accurate prediction of drug-disease association through feature fusion and MLP network. Therefore, the present invention solves the problems of insufficient local feature capture, poor multidimensional feature fusion ability, and poor performance in sparse and unbalanced data caused by static modeling of drug-disease networks in existing methods, and has significant advantages of high accuracy, strong robustness and good generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0096] Figure 1 It is a schematic diagram of a process of a drug repositioning method based on subgraph perception and hybrid graph neural network in one embodiment of the present application;

[0097] Figure 2 This is a schematic diagram of sub-graph extraction in one embodiment of the present application;

[0098] Figure 3 This is a comparison chart of new drug prediction experiments in one embodiment of the present application;

[0099] Figure 4 It is a system flow diagram of a drug relocation system based on subgraph perception and hybrid graph neural network in one embodiment of the present application. DETAILED DESCRIPTION

[0100] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0101] Example 1

[0102] like Figure 1 As shown, the embodiment of the present application discloses a drug repositioning method based on subgraph perception and hybrid graph neural network, including:

[0103] S1. In the drug-disease association network In this paper, the target drug-disease pair is the center, from the target drug-disease pair Extract all drug nodes and disease nodes related to the target node in the neighborhood, as well as the edges between these nodes, to generate the corresponding dynamic drug-disease association subgraph ;

[0104] In some embodiments, the ; Use known drug-disease associations to build a drug-disease association network ,in is the drug node set, Include Different drugs, is the set of disease nodes, Include different diseases; is the edge set, ,in Represents drug node and disease nodes If the relationship between drugs is known and disease If there is a relationship, then the edge The weight of Set to 1, otherwise 0.

[0105] In this embodiment, the dynamic drug-disease association subgraph The target drug-disease pair is centrally processed in the drug-disease association network. Extract the local subgraph consisting of all associated nodes and edges in the neighborhood, such as Figure 2 (a) The target drug -disease As a central node, of =2 The neighbor nodes in the neighborhood have , of The neighbor nodes of , integrate these nodes to construct target drugs -disease A subgraph such as Figure 2 (b) The subgraph retains the topological structure and interaction relationships around the target node to capture local features.

[0106] In order to accurately represent the dynamic drug-disease association subgraph The hierarchical structure of the proposed method starts from the target drug node and the disease node, and is hierarchically labeled according to the shortest path distance from the node to the central node (the shortest path distance to the drug node and the disease node of the target drug-disease pair) to distinguish the hierarchical information of different nodes, while retaining the relative position relationship between the nodes, retaining the topological structure and interaction relationship around the target node, ensuring the clarity of the expression of the subgraph structure features, and can be used to capture local features.

[0107] In order to avoid data leakage in the prediction task, in the dynamic drug-disease association subgraph During the construction of the target drug, direct edges between the target drug and the target disease were deleted (e.g. Figure 3 of and ), ensuring that the model is not distracted by known associations during learning.

[0108] By extracting subgraphs for feature extraction, the network is prevented from being too large and noisy. Compared with the global feature aggregation method used in the full-graph learning model, the efficiency is improved and the key detail features at the subgraph level between specific drug-disease pairs are avoided. These details usually carry important semantic information and local interaction features. Feature extraction through subgraphs allows the model to focus on meaningful interactions.

[0109] S2, using the similarity data between drug nodes and disease nodes, The drug nodes and disease nodes in the , respectively, construct drug subgraphs (drug similarity networks) and disease subgraphs (disease similarity networks) ;

[0110] In some embodiments, the drug subgraph ,in for The node collection in contains All drug nodes in for The edge set in , represents Similarity data between drug nodes in the disease subgraph ),in for The node collection in contains All disease nodes in for The edge set in , represents Similarity data between disease nodes in .

[0111] In some embodiments, based on the chemical structure information (such as SMILES representation) obtained from the DrugBank database, the similarity between all drug nodes is calculated using the Tanimoto similarity formula to generate a drug similarity matrix. , Middle Line Elements of a column Indicates Drug nodes and Drug similarity matrix Represents the similarity data between drug nodes;

[0112] In some embodiments, the phenotypic description of the disease (such as MeSH terms) is obtained from the OMIM disease phenotype database, and the similarity between all disease nodes is obtained using a semantic calculation method based on Jaccard similarity. , thus generating a disease similarity matrix , Middle Line Elements of a column Indicates Disease nodes and The similarity between disease nodes; using the disease similarity matrix Represents similarity data between disease nodes;

[0113] Based on the similarity data between drug nodes and the similarity data between disease nodes, the K nearest neighbor similarity matrix of drugs and diseases is constructed respectively. and , Middle Line Elements of a column and Middle Line Elements of a column The definition is as follows:

[0114] ;

[0115] in, Representative Drug nodes Neighbor set, if drug nodes belong to the K nearest neighbor set, the similarity value remains the original similarity value , otherwise set to 0.

[0116] ;

[0117] in, Representative Disease Node Neighbor set, if disease nodes belong to the K nearest neighbor set, the similarity value remains the original similarity value , otherwise set to 0.

[0118] Get the K nearest neighbor similarity matrix of drugs and diseases and After the matrix, the two are respectively used as and The weight of the corresponding edge in .

[0119] S3, yes The multi-head dynamic graph attention neural network model is used for deep feature extraction to obtain The final feature representation ;

[0120] In some embodiments, in step S3, the multi-head dynamic graph attention neural network model includes a multi-layer neural network.

[0121] In a specific embodiment, the feature extraction network is composed of a three-layer neural network, and the number of attention heads in the three layers is 6. The input dimension of the first layer network is the dimension of the original feature of the node, and the hidden layer dimension is 128; the input dimension of the second layer network is the output dimension of the first layer multiplied by the number of attention heads, and the hidden layer dimension is 64; the input dimension of the third layer is the output dimension of the second layer multiplied by the number of attention heads, and the hidden layer dimension is 1.

[0122] Each layer of the neural network dynamically calculates the attention weight (attention coefficient) between each node and its neighboring nodes through the attention mechanism, thereby extracting the interaction characteristics between nodes; the features of adjacent nodes are weighted and summed to obtain the updated feature representation of the target node, and the ReLU activation function is used for nonlinear activation and the normalization function is used for normalization to enhance the nonlinear ability of feature expression; after the output features of the multi-layer neural network are fused, the trainable normalization weights are learned using the differentiable pooling method, and the node features are grouped and aggregated to obtain The final feature representation In this way, the local structure and global topological information of the subgraph can be preserved, and the local interaction relationship and global topological structure information can be fully captured.

[0123] Through this step, different importance can be assigned to the neighborhood nodes, and feature aggregation is more accurate.

[0124] Specifically, step S3 includes:

[0125] S301, in Layer neural network In an attention head, the node With Node The attention weight is calculated as follows:

[0126] ;

[0127] in, ; Representative Node of The set of neighbor nodes in the neighborhood; and Represents nodes and In step S1, the nodes may be hierarchically labeled according to the shortest path distance from the node to the target drug-disease pair, and then the hierarchical labeling information of the nodes may be ont-hot encoded as the feature vector of the node; For the Layer Neural Network The weight matrix of the attention heads is used to map the input features into a high-dimensional latent space. For the Layer neural network The saliency weight vector of the features connected by the attention heads; and The value of can be randomly initialized and changes as the model learns;

[0128] S302. Based on the attention weight, a multi-head attention mechanism is used to perform weighted summation (dynamic aggregation) of the neighboring node features of the target node. The formula is defined as:

[0129] ;

[0130] in, Representation Node In the The updated feature representation in the layer neural network, Represents the number of attention heads;

[0131] In the above steps, each attention head independently calculates the attention weight and averages the features of adjacent nodes to ensure consistent dimensions;

[0132] S303, yes The ReLU activation function is used for nonlinear activation and the normalization function is used for normalization to enhance the nonlinear ability of feature expression and obtain the node In the Output features of layer neural network ;

[0133] S304, the output features of the multi-layer neural network are integrated to obtain The characteristic matrix ; The feature matrix No. Behavior , ;

[0134] That is to say, Each row in corresponds to the feature vector of a node;

[0135] S305, using the differentiable pooling method to learn the trainable normalized weights, grouping and aggregating the node features, The characteristic matrix and the adjacency matrix Mapped to smaller subgraphs via the following transformation:

[0136] ;

[0137] ;

[0138] in, is a learnable normalized weight matrix used to distribute the original nodes into fewer super nodes, The dimension is , is the original number of nodes, is the number of super nodes after pooling, where the element Represents the original node Assign to super nodes The weight of Initialized randomly and changes as the model learns; for The adjacency matrix describes the connection relationship between nodes, where the elements are The weight of the corresponding edge in ; Represents the feature matrix of the super node, It is the pooled adjacency matrix, describing the relationship between super nodes;

[0139] and constitute The final feature representation .

[0140] By learning trainable normalized weights through differentiable pooling methods and grouping and aggregating node features, the scale of the drug-disease subgraph can be reduced, the receptive field range of the node can be gradually adjusted, and the dilution of the representation by redundant information can be avoided, thereby improving the efficiency and generalization ability of the model.

[0141] S4, based on the fusion of the graph convolution network GCN model and the graph attention network GAT model, and The network performs feature extraction and obtains and The final feature representation and ;

[0142] This step uses the GCN model to aggregate neighborhood information to capture the local structural characteristics of the node, and calculates the attention weight between the node and the neighboring nodes through the GAT model, thereby extracting the global similarity relationship of the nodes; thus, the structural characteristics and similarity relationship of the drug node and the disease node can be extracted;

[0143] In a specific embodiment, the input dimension of the GCN model layer is the dimension of the drug subgraph / disease subgraph, that is, the number of columns of the similarity matrix, and the output dimension is 128. The importance weights between nodes and neighboring nodes are calculated through the graph attention network (GAT model), thereby extracting the global similarity relationship of the nodes. The dimension of the GAT model layer is set to be the same as that of the GCN model layer, with 64 hidden layers.

[0144] In some embodiments, this step includes:

[0145] S401, using the GCN model to aggregate the features of the target node and its adjacent nodes; wherein the GCN model includes layer neural network, where 2. Drug Node and disease nodes In the The feature update formula of the layer neural network is:

[0146] ;

[0147] ;

[0148] in, , Respectively represent Layer Neural Network Drug Node and disease nodes The local structural characteristics, normalized coefficient and Defined as:

[0149] ;

[0150] ;

[0151] in, and Represents drug nodes and drug nodes Drug subgraph The degree in (the sum of in-degree and out-degree); and Represents disease nodes and disease nodes In the disease subgraph The degree in ;

[0152] , In the GCN model, Layer Neural Network for Drug Subgraphs and the disease subgraph The learnable parameter matrix, whose values ​​can be randomly initialized and optimized by the back-propagation algorithm during training;

[0153] Calculated by the above formula and ;

[0154] S402, using the GAT model to capture the interaction between adjacent nodes; wherein the GAT model includes layer neural network, where 2; Each layer of the neural network uses a multi-head attention mechanism to focus on the drug nodes and disease nodes Update the characteristics of the drug node and disease nodes In the GAT model The feature update formula of the layer neural network is defined as follows:

[0155] ;

[0156] ;

[0157] and Drug Node and disease nodes In the GAT model The node features after the layer neural network is updated, is the number of attention heads included in the GAT model, In the GAT model The first layer of the neural network Drug nodes in the attention head and drug nodes The attention weights between In the GAT model The first layer of the neural network Disease nodes in attention heads and disease nodes The attention weights between and In the GAT model, Layer Neural Network for Drug Subgraphs and the disease subgraph The learnable parameter matrix, whose values ​​can be randomly initialized and optimized by the back-propagation algorithm during training;

[0158] Calculated by the above formula and ;

[0159] S403, weighted fusion is performed on the outputs of the GCN model and the GAT model to generate the final feature representation of the target node; the fusion process is defined as:

[0160] ;

[0161] ;

[0162] in, and Represents drug nodes and disease nodes The final feature representation is, It is the fusion ratio hyperparameter, which is used to adjust the relative contribution of the output of the GCN model and the GAT model;

[0163] By drug subgraph The final feature representation of all drug nodes in constitutes the drug subgraph The final feature representation ;

[0164] Disease subgraph The final feature representation of all disease nodes in constitutes the disease subgraph The final feature representation .

[0165] This step can improve the ability to identify implicit patterns in drug-disease associations, thereby improving the overall expressiveness of the model.

[0166] S5. , and The final feature representation , , Fusion to build a unified feature representation ;Will The drug-disease association prediction model based on the multi-layer perceptron MLP network is input, and the drug-disease association prediction model outputs the prediction result of the association relationship of the target drug-disease pair.

[0167] Will , and The final feature representation , , The fusion is performed to obtain a unified feature representation that combines the characteristics of drug and disease nodes. The unified feature representation is transformed nonlinearly through MLP to learn the complex interaction relationship between drug and disease nodes.

[0168] In some embodiments, , and The final feature representation , Fusion to build a unified feature representation , the feature fusion process is defined as follows:

[0169] ;

[0170] in, Represents a splicing operation.

[0171] In some embodiments, the output layer of the drug-disease association prediction model predicts the drug node by the following formula and disease nodes The probability of association between them, the greater the probability, the more likely the association is:

[0172] ;

[0173] in, , , and are weight and coefficient parameters, their values ​​can be randomly initialized and they are optimized by the back-propagation algorithm during the training process.

[0174] In some embodiments, the multi-head dynamic graph attention neural network model in step S3, the graph convolutional network GCN model and the graph attention network GAT model in S4, and the drug-disease association prediction model based on the multi-layer perceptron MLP network in step S5 are obtained by training and optimization based on the known target drug-disease pair association relationship.

[0175] The cross entropy loss function is defined as follows:

[0176] ;

[0177] in, Drug Node and disease nodes The true label of the correlation relationship between Predicted drug nodes and disease nodes The probability of association between is the total number of samples. Among them, the samples are target drug-disease pairs with known associations.

[0178] In some embodiments, a regularization term is introduced to improve the generalization ability of the model, including adding a regularization term to the objective function, which is defined as:

[0179] ;

[0180] in, represents the regularization coefficient, Represents the learnable parameters of the k-th layer of the model.

[0181] It should be understood that the above numbers S1 to S5, S301 to S305, and S401 to S403 are only used to distinguish and facilitate the expression of different steps, and do not necessarily constitute a limitation on the execution order of the steps.

[0182] In the examples of this application, data is collected: experiments are conducted on three data sets, namely Fdataset, Cdataset and Lrssl. Drug information comes from the DrugBank database, and disease information comes from the OMIM database. Each data set includes a drug and disease similarity matrix, a drug-disease association matrix, and drug and disease IDs. The drug ID is taken from DrugBank, and the disease ID is taken from OMIM. Among them, Fdataset contains 593 drugs, 313 diseases, and 1933 drug-disease associations, with a sparsity of 0.0104. The three data sets are described in detail in Table 1 below:

[0183]

[0184] In this embodiment, a 10-fold cross validation method is used to evaluate each data set. In addition, in order to prevent overfitting and improve the accuracy of the evaluation, an equal number of negative samples are systematically included in the training and testing process to match the number of positive samples to form a 1:1 ratio. AUROC and AUPRC are used as the main indicators for the comprehensive evaluation of the overall performance of the model.

[0185] Table 2 shows a detailed comparison of the method provided in the embodiment of the present application and six benchmark models NIMCGCN, DRWBNCF, iDrug, PSGCN, DRAGNN and WIGRL on three benchmark datasets. The benchmark models, such as the DRWBNCF model, use weighted bilinear operations to integrate drug-disease associations and similarity networks; the PSGCN model extracts subgraphs of drug-disease pairs and uses a hierarchical attention mechanism to capture multi-scale features. It can be seen that the method provided in the embodiment of the present application is superior to other models in all indicators. It is worth noting that on the sparse dataset Lrssl, the method provided in the embodiment of the present application achieves an AUROC of 0.9655, which is significantly higher than the second best PSGCN of 0.942, highlighting its ability to effectively process sparse data.

[0186]

[0187] In this example, in order to solve the cold start problem and evaluate the ability to predict the indications of new drugs, a leave-one-out cross-validation experiment was performed. This experiment excluded all known drug-disease associations and used them as the test set. The associations of the five most similar drugs determined by the similarity score were used to update the drug-disease graph, while the remaining confirmed associations were used as the training set. The results are shown in Figure 2. Figure 3 As shown, the AUROC of the present invention on Fdataset is 0.9237, which is significantly better than other baseline models.

[0188] Embodiment 2

[0189] like Figure 4As shown, the embodiment of the present application is a drug relocation system based on subgraph perception and hybrid graph neural network, including a data input layer, a subgraph extraction layer, a feature extraction layer, a multi-head dynamic graph attention neural network model layer, a GCN model and a GAT model fusion neural network layer, and a feature fusion and prediction layer;

[0190] The data input layer is used to collect and integrate data on drugs, diseases and their relationships;

[0191] In some embodiments, the data of drugs, diseases and their associations include chemical structure information of drugs (such as SMILES representation), phenotypic descriptions of diseases (such as MeSH terms, and association data between drugs and diseases; these data are derived from public databases (such as DrugBank and OMIM) and are standardized into a unified structured form through a data input layer; the data input layer constructs a drug node set and a disease node set, and connects the drug nodes and the disease nodes through association relationships to form a drug-disease association network;

[0192] The subgraph extraction layer is used to extract the drug-disease association network Extracting dynamic drug-disease association subgraphs from , to capture the local structural features and interactive relationships of the target drug-disease pair; specifically, the subgraph extraction layer is used in the drug-disease association network In this paper, the target drug-disease pair is the center, from the target drug-disease pair In the neighborhood, all drug nodes and disease nodes related to the target node, as well as the edges between these nodes, are extracted to generate a dynamic drug-disease association subgraph ;

[0193] In some embodiments, the subgraph extraction layer uses a distance coding method to hierarchically label nodes according to the shortest path distance from each node to the target drug or disease node to clarify the hierarchical relationship of the nodes in the subgraph;

[0194] The feature extraction layer is used to utilize the similarity data between drug nodes and the similarity data between disease nodes to The drug nodes and disease nodes in the , respectively, construct drug subgraphs (drug similarity networks) and disease subgraphs (disease similarity networks) ;

[0195] Specifically, the feature extraction layer can calculate the similarity between drug nodes by analyzing the chemical structure information of the drugs to generate a drug similarity matrix; at the same time, the phenotypic information of the disease is used to calculate the similarity between disease nodes to generate a disease similarity matrix; further, the feature extraction layer is used to calculate the similarity between disease nodes to generate a disease similarity matrix. Neighbor drug similarity matrix and Based on the neighbor disease similarity matrix, drug subgraphs are constructed separately and the disease subgraph , where the nodes of the network are dynamic drug-disease association subgraphs The drug or disease nodes in the ,edges are similarity relationships between nodes;

[0196] The multi-head dynamic graph attention neural network model layer is used to analyze the dynamic drug-disease association subgraph Perform deep feature extraction to obtain The final feature representation ;

[0197] This layer uses a multi-head attention mechanism to independently calculate the attention weights between nodes and neighboring nodes in each layer, and performs weighted aggregation on the features of neighboring nodes to generate node representations in different semantic spaces; and uses the normalized weights that can be trained using differentiable pooling technology to group and aggregate node features to obtain The final feature representation , thereby retaining key local interaction characteristics and global structural information while compressing the subgraph size;

[0198] The GCN model and the GAT model fusion neural network layer are used to fuse the graph convolution network GCN model and the graph attention network GAT model respectively. and The network performs feature extraction and obtains and The final feature representation and ;

[0199] Based on drug subgraph and the disease subgraph The structural characteristics and similarity relationships of nodes can be extracted. This layer aggregates the features of neighbor nodes through the graph convolutional network (GCN model) to capture the local structural information of drug nodes and disease nodes. At the same time, the graph attention network (GAT model) is used to calculate the attention weights between the node and its neighbor nodes to extract the global similarity relationship. Finally, the output results of the GCN model and the GAT model are weighted and fused to generate a comprehensive feature representation.

[0200] The feature fusion and prediction layer is used to , and The final feature representation , , Fusion to build a unified feature representation ;Will The drug-disease association prediction model based on the multi-layer perceptron MLP network is input, and the drug-disease association prediction model outputs the prediction result of the association relationship of the target drug-disease pair.

[0201] The system realizes drug relocation based on the method described in the first embodiment.

[0202] Embodiment 3

[0203] This embodiment provides an electronic device, including: a memory and a processor;

[0204] The memory is used to store computer programs;

[0205] The processor is used to call the computer program to execute the method as described in the first embodiment.

[0206] Embodiment 4

[0207] This embodiment provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed on an electronic device, the electronic device implements the method described in the first embodiment.

[0208] Embodiment 5

[0209] This embodiment provides a computer program product, including a computer program. When the computer program is executed on an electronic device, the electronic device implements the method described in the first embodiment.

[0210] The specific implementation methods of a system, electronic device, computer-readable storage medium, and computer program product provided in the embodiments of the present application can refer to the specific embodiments of the above-mentioned method and will not be repeated here.

[0211] Obviously, those skilled in the art should understand that the above-mentioned units or steps of the present application can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the present application is not limited to any specific combination of hardware and software.

[0212] Unless otherwise expressly stated, the terms and expressions used in this specification should be interpreted in accordance with the meanings generally understood by those of ordinary skill in the art. The specific embodiments described herein are intended to explain the technical principles and methods of the present invention in detail, but should not be regarded as a limiting description of the present invention. In order to ensure that the expression is concise and clear, not every possible combination of technical features is elaborated in detail, but as long as the combination of these technical features does not conflict logically, they should be considered to fall within the scope of the content covered by this specification.

[0213] After reading this specification, those skilled in the art may make various modifications, adjustments or equivalent changes to the implementation methods without departing from the basic spirit and technical core of the present invention. These modifications and changes, including but not limited to changes in the structural form and step sequence shown in the drawings, shall be regarded as extensions of the present invention as long as the results are still within the scope of protection of the claims. In addition, the attached claims not only clearly define the scope of protection of the present invention, but also cover equivalent forms or reasonable extensions of these claims.

Claims

1. A drug repositioning method based on subgraph perception and hybrid graph neural network, characterized in that: include: S1. In the drug-disease association network In this paper, the target drug-disease pair is the center, from the target drug-disease pair Extract all drug nodes and disease nodes related to the target node in the neighborhood, as well as the edges between these nodes, to generate the corresponding dynamic drug-disease association subgraph ; S2, using the similarity data between drug nodes and disease nodes, The drug nodes and disease nodes in the and the disease subgraph ; S3, yes The multi-head dynamic graph attention neural network model is used for deep feature extraction to obtain The final feature representation ; S4, based on the fusion of the graph convolution network GCN model and the graph attention network GAT model, and The network performs feature extraction and obtains and The final feature representation and ; S5. , and The final feature representation , , Fusion to build a unified feature representation ;Will Input a drug-disease association prediction model based on a multi-layer perceptron MLP network, and output the prediction result of the association relationship of the target drug-disease pair through the drug-disease association prediction model; The multi-head dynamic graph attention neural network model in step S3, the graph convolutional network GCN model and the graph attention network GAT model in step S4, and the drug-disease association prediction model based on the multi-layer perceptron MLP network in step S5 are obtained through training and optimization based on the known target drug-disease pair association relationship.

2. The method according to claim 1, characterized in that In step S1, a drug-disease association network is constructed using known drug-disease association relationships. ,in is the drug node set, Include Different drugs, is the set of disease nodes, Include different diseases; is the edge set, ,in Represents drug node and disease nodes If the relationship between drugs is known and disease If there is a relationship, then the edge The weight of Set to 1, otherwise 0.

3. The method according to claim 1, characterized in that In step S2, the drug subgraph ,in for The node collection in contains All drug nodes in for The edge set in , represents Similarity data between drug nodes in the disease subgraph ),in for The node collection in contains All disease nodes in for The edge set in , represents Similarity data between disease nodes in ; In step S2, based on the chemical structure information obtained from the DrugBank database, the similarity between all drug nodes is calculated to generate a drug similarity matrix. , Middle Line Elements of a column Indicates Drug nodes and The similarity between drug nodes; Using the drug similarity matrix Represents the similarity data between drug nodes; In step S2, the phenotypic description of the disease is obtained from the OMIM disease phenotype database, and the similarity between all disease nodes is calculated. , thus generating a disease similarity matrix , Middle Line Elements of a column Indicates Disease nodes and Disease similarity matrix Represents similarity data between disease nodes; Based on the similarity data between drug nodes and the similarity data between disease nodes, the K nearest neighbor similarity matrix of drugs and diseases is constructed respectively. and , Middle Line Elements of a column and Middle Line Elements of a column The definition is as follows: ； in, Representative Drug nodes Neighbor set, if drug nodes belong to the K nearest neighbor set, the similarity value remains the original similarity value , otherwise set to 0; ； in, Representative Disease Node Neighbor set, if disease nodes belong to the K nearest neighbor set, the similarity value remains the original similarity value , otherwise set to 0; Get the K nearest neighbor similarity matrix of drugs and diseases and After the matrix, the two are respectively used as and The weight of the corresponding edge in .

4. The method according to claim 1, characterized in that: In step S3, the multi-head dynamic graph attention neural network model includes a multi-layer neural network; The step S3 comprises: S301, in Layer neural network In an attention head, the node With Node The attention weight is calculated as follows: ； in, , Representative Node of The set of neighbor nodes in the neighborhood, and Represents nodes and The eigenvector of For the Layer Neural Network The weight matrix of the attention heads is used to map the input features into a high-dimensional latent space. For the Layer neural network The saliency weight vector of the features connected by the attention heads; S302. Based on the attention weight, a multi-head attention mechanism is used to perform weighted summation (dynamic aggregation) of the features of the neighboring nodes of the target node. The formula is defined as: ； in, Representation Node In the The updated feature representation in the layer neural network, Represents the number of attention heads; S303, yes The ReLU activation function is used for nonlinear activation and the normalization function is used for normalization to enhance the nonlinear ability of feature expression and obtain the node In the Output features of layer neural network ; S304, the output features of the multi-layer neural network are integrated to obtain The characteristic matrix ; The feature matrix No. Behavior , ; S305, using a differentiable pooling method to learn trainable normalized weights, grouping and aggregating node features to reduce the size of the drug-disease subgraph and gradually expand the receptive field range of the node; The characteristic matrix and the adjacency matrix Mapped to smaller subgraphs via the following transformation: ； ； in, is a learnable normalized weight matrix used to distribute the original nodes into fewer super nodes, The dimension is , is the original number of nodes, is the number of super nodes after pooling, where the element Represents the original node Assign to super nodes The weight of for The adjacency matrix of The weight of the corresponding edge in ; Represents the feature matrix of the super node, is the adjacency matrix after pooling; and constitute The final feature representation .

5. The method according to claim 1, characterized in that The step S4 comprises: S401, using the GCN model to aggregate the features of the target node and its adjacent nodes; wherein the GCN model includes layer neural network, where 2. Drug Node and disease nodes In the The feature update formula of the layer neural network is: ； ； in, , Respectively represent Layer Neural Network Drug Node and disease nodes The local structural characteristics, normalized coefficient and Defined as: ； ； in, and Represents drug nodes and drug nodes In the drug subgraph The degree in ; and Represents disease nodes and disease nodes In the disease subgraph The degree in ; , In the GCN model, Layer Neural Network for Drug Subgraphs and the disease subgraph The learnable parameter matrix of Calculated by the above formula and ; S402, using the GAT model to capture the interaction between adjacent nodes; wherein the GAT model includes layer neural network, where 2; Each layer of the neural network uses a multi-head attention mechanism to focus on the drug nodes and disease nodes Update the characteristics of the drug node and disease nodes In the GAT model The feature update formula of the layer neural network is defined as follows: ； ； and Drug Node and disease nodes In the GAT model The node features after the layer neural network is updated, is the number of attention heads included in the GAT model, In the GAT model The first layer of the neural network Drug nodes in the attention head and drug nodes The attention weights between In the GAT model The first layer of the neural network Disease nodes in attention heads and disease nodes The attention weights between and In the GAT model, Layer Neural Network for Drug Subgraphs and the disease subgraph The learnable parameter matrix of Calculated by the above formula and ; S403, weighted fusion is performed on the outputs of the GCN model and the GAT model to generate the final feature representation of the target node; the fusion process is defined as: ； ； in, and Represents drug nodes and disease nodes The final feature representation of It is the fusion ratio hyperparameter, which is used to adjust the relative contribution of the output of the GCN model and the GAT model; By drug subgraph The final feature representation of all drug nodes in constitutes the drug subgraph The final feature representation ; By disease subgraph The final feature representation of all disease nodes in constitutes the disease subgraph The final feature representation .

6. The method according to claim 1, characterized in that The The drug-disease association prediction model based on the multi-layer perceptron MLP network is input, and the drug-disease association prediction model outputs the prediction results of the association relationship of the target drug-disease pair, including: The output layer of the drug-disease association prediction model predicts the drug node by the following formula and disease nodes The probability of association between: ； in, , , and are weight and coefficient parameters.

7. A drug relocation system based on subgraph perception and hybrid graph neural network, characterized in that: It includes data input layer, sub-graph extraction layer, feature extraction layer, multi-head dynamic graph attention neural network model layer, GCN model and GAT model fusion neural network layer, feature fusion and prediction layer; The data input layer is used to collect and integrate data on drugs, diseases and their associations to form a drug-disease association network. ; The subgraph extraction layer is used in the drug-disease association network In this paper, the target drug-disease pair is the center, from the target drug-disease pair In the neighborhood, all drug nodes and disease nodes related to the target node, as well as the edges between these nodes, are extracted to generate a dynamic drug-disease association subgraph ; The feature extraction layer is used to utilize the similarity data between drug nodes and the similarity data between disease nodes to The drug nodes and disease nodes in the and the disease subgraph ; The multi-head dynamic graph attention neural network model layer is used to analyze the dynamic drug-disease association subgraph Perform deep feature extraction to obtain The final feature representation ; The GCN model and the GAT model fusion neural network layer are used to fuse the graph convolution network GCN model and the graph attention network GAT model respectively. and The network performs feature extraction and obtains and The final feature representation and ; The feature fusion and prediction layer is used to , and The final feature representation , , Fusion to build a unified feature representation ;Will Input a drug-disease association prediction model based on a multi-layer perceptron MLP network, and output the prediction result of the association relationship of the target drug-disease pair through the drug-disease association prediction model; The system achieves drug relocation based on the method according to any one of claims 1 to 6.

8. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is configured to call the computer program to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed on an electronic device, the electronic device implements the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed on an electronic device, the electronic device implements the method according to any one of claims 1 to 6.

Images