Drug-disease association prediction method, device and electronic device based on hypergraph

Abstract

The present invention provides a hypergraph-based drug-disease association prediction method, device and electronic device, which relates to the field of bioinformatics, including: obtaining a drug hypergraph and a disease hypergraph; based on a hierarchical contrastive learning algorithm, inputting the drug hypergraph and the disease hypergraph into a target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix; performing interactive information aggregation through the initial drug feature matrix and the initial disease feature matrix to obtain a target drug feature matrix and a target disease feature matrix; performing association prediction through the target drug feature matrix and the target disease feature matrix to obtain a drug-disease potential association matrix. The present invention realizes accurate prediction of drug-disease potential association.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a hypergraph-based drug-disease association prediction method, device and electronic equipment. Background Art

[0002] With the continuous development of computers, the cost and time requirements of traditional drug development are constantly increasing. In the process of drug development, machine learning technology is often used to predict potential drug-disease associations, which helps to identify candidate drugs with clinical validation potential and significantly reduce the time and cost required to bring effective treatments to market.

[0003] Currently, the drug-disease association is mainly obtained through simple drug-disease association graphs or similarity graphs, but the drug-disease association is highly sparse. The prediction method for drug-disease association is difficult to effectively handle the complexity of data association, cannot meet actual needs, and cannot make accurate predictions. Summary of the invention

[0004] Problems to be solved by the present invention are to solve at least one of the above problems.

[0005] To solve the above problems, the present invention provides a hypergraph-based drug-disease association prediction method, device and electronic device.

[0006] In a first aspect, the present invention provides a hypergraph-based drug-disease association prediction method, comprising: Obtain drug hypergraphs and disease hypergraphs; Based on a hierarchical contrastive learning algorithm, the drug hypergraph and the disease hypergraph are input into a target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix; Obtaining a target drug feature matrix and a target disease feature matrix by interactively aggregating the initial drug feature matrix and the initial disease feature matrix; The drug-disease potential association matrix is ​​obtained by performing association prediction on the target drug feature matrix and the target disease feature matrix.

[0007] Optionally, obtaining a drug hypergraph and a disease hypergraph includes: Acquire a drug similarity network and a disease similarity network, wherein the drug similarity network includes a drug function similarity network and a drug association similarity network, and the disease similarity network includes a disease semantic similarity hypergraph network and a disease association similarity hypergraph network; A drug hyperedge set and a disease hyperedge set are obtained through the drug similarity network and the disease similarity network, wherein the drug hyperedge set includes a drug function similarity hyperedge group and a drug association similarity hyperedge group, and the disease hyperedge set includes a disease semantic similarity hyperedge group and a disease association similarity hyperedge group; Among them, the drug function similar super marginal group is: , in, is the drug function similarity hyperedge group, k is the number of adjacent drug nodes, is a set of nodes with similar drug functions, v is a vertex of a given drug, is the drug node set; Among them, the drug-associated similarity super-marginal groups are: , in, for the drug-associated similarity super-marginal group, A set of similar nodes associated with drugs; The drug hypergraph and the disease hypergraph are obtained according to the drug hyperedge set and the disease hyperedge set.

[0008] Optionally, obtaining the drug hypergraph and the disease hypergraph according to the drug hyperedge set and the disease hyperedge set includes: Obtaining a drug adjacency matrix and a disease adjacency matrix respectively through the drug hyperedge set and the disease hyperedge set; Obtaining the drug hypergraph and the disease hypergraph according to the drug adjacency matrix and the disease adjacency matrix respectively; Wherein, the drug hypergraph is: , in, is the drug hypergraph, is the drug adjacency matrix.

[0009] Optionally, the target hypergraph convolution model includes a plurality of sequentially connected hypergraph convolution layers, and the step of inputting the drug hypergraph and the disease hypergraph into the target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix includes: Inputting the drug hypergraph and the disease hypergraph into the first hypergraph convolution layer of a plurality of sequentially connected hypergraph convolution layers respectively, extracting features in sequence through the plurality of hypergraph convolution layers to obtain the initial drug feature matrix and the initial disease feature matrix, wherein features are extracted in each of the hypergraph convolution layers by an information propagation algorithm; Wherein, the information propagation algorithm is: , in, is the output feature matrix of the l+1th layer, is any nonlinear activation function, is the diagonal matrix from vertices to hyperedges, H is the adjacency matrix of the hypergraph, and W is the matrix that records the hyperedge weights. is the diagonal matrix from hyperedges to vertices, is the transposed adjacency matrix of the hypergraph, is the preset output feature matrix of the lth layer, is the trainable parameter matrix.

[0010] Optionally, the process of constructing the target hypergraph convolutional model includes: Get the initial hypergraph convolution model; Inputting the training data into the initial hypergraph convolutional model to obtain a prediction result, and calculating the layer-wise contrastive learning loss according to the prediction result; Among them, the layered contrastive learning loss is: , in, is the layer-wise contrastive learning loss, is the cross entropy loss, Contribute parameters to adjust the loss, is the contrast loss of each layer; The model parameters of the initial hypergraph convolutional model are optimized by the hierarchical contrastive learning loss until the target hypergraph convolutional model is obtained.

[0011] Optionally, the interactive information aggregation of the initial drug feature matrix and the initial disease feature matrix to obtain a target drug feature matrix and a target disease feature matrix includes: Obtaining the target drug feature matrix by interactively aggregating information from the initial drug feature matrix; Among them, the target drug feature matrix is: , in, is the target drug feature matrix, is a cross-attention module for improving representation quality, is the initial drug feature matrix; Obtaining the target disease feature matrix by interactively aggregating information from the initial disease feature matrix; Among them, the target disease feature matrix is: , in, is the target disease feature matrix, is the initial disease feature matrix.

[0012] Optionally, the obtaining of a drug-disease potential association matrix by performing association prediction on the target drug feature matrix and the target disease feature matrix includes: The drug-disease potential association matrix is ​​obtained by performing association prediction on the target drug feature matrix and the target disease feature matrix; Wherein, the drug-disease potential association matrix is: , in, is the drug-disease potential association matrix, is a function that maps predicted values ​​to a preset range, is the target drug feature matrix, is the target disease feature matrix, is the trainable parameter matrix.

[0013] In a second aspect, the present invention provides a hypergraph-based drug-disease association prediction device, comprising: a hypergraph acquisition module, for acquiring a drug hypergraph and a disease hypergraph; An initial feature matrix acquisition module, used for inputting the drug hypergraph and the disease hypergraph into a target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix based on a hierarchical contrastive learning algorithm; A target feature matrix acquisition module, used for interactively aggregating information from the initial drug feature matrix and the initial disease feature matrix to obtain a target drug feature matrix and a target disease feature matrix; The drug-disease potential association matrix acquisition module is used to obtain the drug-disease potential association matrix by performing association prediction on the target drug feature matrix and the target disease feature matrix.

[0014] In a third aspect, the present invention provides an electronic device, including a memory and a processor; The memory is used to store computer programs; The processor is used to implement the hypergraph-based drug-disease association prediction method as described in the first aspect when executing the computer program.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the hypergraph-based drug-disease association prediction method as described in the first aspect is implemented.

[0016] The beneficial effects of the hypergraph-based drug-disease association prediction method, device and electronic device of the present invention are: using the existing drug-disease associations to construct drug hypergraphs and disease hypergraphs, complex connections can be better represented. Applying the target hypergraph convolution model to perform node feature learning to obtain the initial drug feature matrix and the initial disease feature matrix can provide more comprehensive information. Capturing the collaborative relationship between different modalities based on the hierarchical contrastive learning algorithm prevents the introduction of redundant information due to excessive attention to global information. Interactive information aggregation is performed to obtain the target drug feature matrix and the target disease feature matrix, further enhancing the model's ability to capture interactive information. Association prediction is performed through the target drug feature matrix and the target disease feature matrix to obtain the drug-disease potential association matrix, which can accurately predict the potential drug-disease association. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a process of a hypergraph-based drug-disease association prediction method according to an embodiment of the present invention; Figure 2 Schematic diagram of the structure of a drug-disease association prediction device based on a hypergraph according to an embodiment of the present invention; Figure 3 The figure is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be interpreted as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not intended to limit the scope of protection of the present invention.

[0019] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0020] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to"; the term "based on" means "based at least in part on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0021] It should be noted that the modifications of "one" and "plurality" mentioned in the present invention are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0022] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes, and are not used to limit the scope of these messages or information.

[0023] In view of the problems existing in the above-mentioned related technologies, this embodiment provides a hypergraph-based drug-disease association prediction method, device and electronic device.

[0024] like Figure 1 As shown, a hypergraph-based drug-disease association prediction method provided by an embodiment of the present invention includes: Step 110, obtaining a drug hypergraph and a disease hypergraph.

[0025] Specifically, a hypergraph is a generalized form of a graph. In a traditional graph, each edge can only connect two vertices, but in a hypergraph, a hyperedge can connect any number of vertices. A hypergraph allows one edge to connect multiple vertices, forming a more complex topological structure.

[0026] Step 120: Based on a hierarchical contrastive learning algorithm, the drug hypergraph and the disease hypergraph are input into a target hypergraph convolutional model to obtain an initial drug feature matrix and an initial disease feature matrix.

[0027] Specifically, the target hypergraph convolutional model can gradually aggregate higher-order neighbor information by stacking multiple hypergraph convolutional layers, thereby capturing more complex dependencies. The output of each layer is used as the input of the next layer to gradually update the feature vector of the node. Hierarchical contrastive learning is a method used in deep learning to generate high-quality feature representations. In particular, when processing hierarchical data, contrastive learning at different levels can improve the representation and generalization capabilities of the model.

[0028] Step 130 , interactively aggregate the initial drug feature matrix and the initial disease feature matrix to obtain a target drug feature matrix and a target disease feature matrix.

[0029] Specifically, a cross-attention mechanism is used for interactive information aggregation, which is first modally spliced ​​and then input into the self-attention mechanism, so that the multimodal data of drugs and diseases can exchange information in mutual attention to obtain the target drug feature matrix and the target disease feature matrix.

[0030] Step 140: performing association prediction on the target drug feature matrix and the target disease feature matrix to obtain a drug-disease potential association matrix.

[0031] Specifically, the target drug feature matrix and the target disease feature matrix are input into the bilinear decoder for association prediction to obtain the drug-disease potential association matrix. The bilinear decoder predicts the association between drugs and diseases by calculating the bilinear product of their potential representations.

[0032] In this embodiment, the existing drug-disease associations are used to construct drug hypergraphs and disease hypergraphs, which can better represent complex connections. The target hypergraph convolution model is applied to perform node feature learning to obtain the initial drug feature matrix and the initial disease feature matrix, which can provide more comprehensive information. The collaborative relationship between different modalities is captured based on the hierarchical contrastive learning algorithm to prevent the introduction of redundant information due to excessive attention to global information. Interactive information aggregation is performed to obtain the target drug feature matrix and the target disease feature matrix, further enhancing the model's ability to capture interactive information. The drug-disease potential association matrix is ​​obtained by association prediction through the target drug feature matrix and the target disease feature matrix, which can accurately predict the potential drug-disease association.

[0033] Optionally, obtaining a drug hypergraph and a disease hypergraph includes: Acquire a drug similarity network and a disease similarity network, wherein the drug similarity network includes a drug function similarity network and a drug association similarity network, and the disease similarity network includes a disease semantic similarity hypergraph network and a disease association similarity hypergraph network; A drug hyperedge set and a disease hyperedge set are obtained through the drug similarity network and the disease similarity network, wherein the drug hyperedge set includes a drug function similarity hyperedge group and a drug association similarity hyperedge group, and the disease hyperedge set includes a disease semantic similarity hyperedge group and a disease association similarity hyperedge group; Among them, the drug function similar super marginal group is: , in, is the drug function similarity hyperedge group, k is the number of adjacent drug nodes, is a set of nodes with similar drug functions, v is a vertex of a given drug, is the drug node set; Among them, the drug-associated similarity super-marginal groups are: , in, for the drug-associated similarity super-marginal group, A set of similar nodes associated with drugs; The drug hypergraph and the disease hypergraph are obtained according to the drug hyperedge set and the disease hyperedge set.

[0034] Specifically, the drug function similarity network and the drug association similarity network are two different types of drug networks, and their main differences lie in their construction basis and the type of relationship between drugs reflected. The drug function similarity network is used to measure the similarity between drugs based on the function or biological characteristics of the drugs. Common functions include drug targets, enzymes, drug interactions, metabolic pathways and substructures. The drug association similarity network is used to measure the similarity between drugs based on the association between drugs and diseases. These associations are usually established through clinical use data of drugs, clinical trial results of drugs, or known relationships between drugs and diseases. The disease semantic similarity hypergraph network is used to calculate the semantic similarity between diseases based on the directed acyclic graph structure obtained from the MeSH descriptor of the disease. The disease association similarity hypergraph network is used to measure the similarity between diseases based on the association between drugs and diseases.

[0035] In the process of constructing the drug function similarity network, different types of feature vectors are constructed for each drug, and based on these features, the Jaccard index similarity measurement method is used to calculate the similarity between drugs. The Jaccard index is a statistic used to measure the similarity between two sets. It evaluates their similarity by calculating the ratio of the intersection size of two sets to the union size of the two sets. Among them, the similarity between the drugs is: , in, for The elements in The number of cases where all corresponding elements in are equal to 1, for Elements or The number of cases where the corresponding element of is equal to 1, For the similarity between drugs, and These are the corresponding characteristics of the two drugs.

[0036] The semantic similarity between diseases is calculated using the directed acyclic graph structure obtained from the MeSH descriptors of the diseases in the process of constructing the disease semantic similarity network. For disease d, we use To represent its hierarchical relationship, is the node set containing disease d and all its ancestor nodes, It represents the set of direct connections from the parent node to the child node. Based on this DAG structure, the contribution of node n to the semantic value of disease d in DAG(d) is: , in, is the contribution factor and is set to 0.5 by default. Child(n) is the set of child nodes of node n. is the contribution of node n to the semantic value of disease d in DAG(d).

[0037] The semantic value of disease d is defined as .in, and The semantic similarity between two diseases is: , in, is the semantic similarity, is d i The node set containing disease d and all its ancestor nodes, is d j The node set containing disease d and all its ancestor nodes, is d i The contribution of middle node n to the semantic value of disease d in DAG(d), is d j The contribution of middle node n to the semantic value of disease d in DAG(d), is d i The semantic value of the disease d in , is d j The semantic value of the disease d in .

[0038] The cosine similarity is used to construct the drug association similarity network and the disease association similarity network. The calculation is as follows: , Where ‖∙‖ is the L2 norm, , are the vectors of drug i and drug j respectively.

[0039] Similarly, according to , we can calculate the association similarity between disease i and disease j as : , in, , are the vectors of disease i and disease j respectively.

[0040] In some more specific embodiments, a drug vertex v is given as a centroid through the drug similarity network and the disease similarity network, respectively, and hyperedges are constructed by connecting its K nearest neighboring drug nodes to obtain a drug hyperedge set and a disease hyperedge set.

[0041] In this optional embodiment, firstly, various similar modes of drugs and diseases are used to construct a drug function similarity network and a disease semantic similarity hypergraph network respectively. Based on the known drug-disease associations, a drug association similarity network and a disease association similarity hypergraph network are constructed, which can integrate drug and disease similarity information from different sources and types.

[0042] Optionally, obtaining the drug hypergraph and the disease hypergraph according to the drug hyperedge set and the disease hyperedge set includes: Obtaining a drug adjacency matrix and a disease adjacency matrix respectively through the drug hyperedge set and the disease hyperedge set; Obtaining the drug hypergraph and the disease hypergraph according to the drug adjacency matrix and the disease adjacency matrix respectively; Wherein, the drug hypergraph is: , in, is the drug hypergraph, is the drug adjacency matrix.

[0043] Specifically, the drug hyperedge set includes a drug function similarity hyperedge group and a drug association similarity hyperedge group. , and obtain two drug association matrices ,in, , M k is the number of hyperedges, n r is the number of drugs, rf is the drug function similarity mode, and ra is the drug association similarity mode. We connect the two association matrices to construct the drug hypergraph Following the same steps, we can construct the disease association matrix ,in, , for the disease hypergraph G d , n d The number of side effects.

[0044] In this optional embodiment, a drug hypergraph and a disease hypergraph are obtained based on the drug adjacency matrix and the disease adjacency matrix. The hypergraph network can effectively capture and represent the many-to-many relationship between drugs and diseases, as well as the high-order associations between them. Compared with the traditional binary relationship network, the hypergraph can provide a more detailed and complex network structure, which helps to reveal potential drug-disease associations.

[0045] Optionally, the target hypergraph convolution model includes a plurality of sequentially connected hypergraph convolution layers, and the step of inputting the drug hypergraph and the disease hypergraph into the target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix includes: Inputting the drug hypergraph and the disease hypergraph into the first hypergraph convolution layer of a plurality of sequentially connected hypergraph convolution layers respectively, extracting features in sequence through the plurality of hypergraph convolution layers to obtain the initial drug feature matrix and the initial disease feature matrix, wherein features are extracted in each of the hypergraph convolution layers by an information propagation algorithm; Wherein, the information propagation algorithm is: , in, is the output feature matrix of the l+1th layer, is any nonlinear activation function, is the diagonal matrix from vertices to hyperedges, H is the adjacency matrix of the hypergraph, and W is the matrix that records the hyperedge weights. is the diagonal matrix from hyperedges to vertices, is the transposed adjacency matrix of the hypergraph, is the preset output feature matrix of the lth layer, is the trainable parameter matrix.

[0046] Specifically, As the input feature set of the lth layer, the information propagation algorithm is: , in, is the output feature matrix of the l+1th layer, is any nonlinear activation function, is the diagonal matrix from vertices to hyperedges, H is the adjacency matrix of the hypergraph, and W is the matrix that records the hyperedge weights. is the diagonal matrix from hyperedges to vertices, is the transposed adjacency matrix of the hypergraph, is the preset output feature matrix of the lth layer, is the trainable parameter matrix.

[0047] in, , , Calculate the sum of each column of the matrix, Used to obtain a diagonal matrix.

[0048] In this optional embodiment, the drug hypergraph and the disease hypergraph are respectively input into the current hypergraph convolution layer, and the hypergraph convolution model can handle more complex relationship structures. At the same time, the information propagation algorithm is used to update the features to obtain the current drug information propagation features and the current disease information propagation features. By transmitting and aggregating node information in the network to generate a feature representation of the node, the node information can be transmitted and aggregated more efficiently.

[0049] Optionally, the process of constructing the target hypergraph convolutional model includes: Get the initial hypergraph convolution model; Inputting the training data into the initial hypergraph convolutional model to obtain a prediction result, and calculating the layer-wise contrastive learning loss according to the prediction result; Among them, the layered contrastive learning loss is: , in, is the layer-wise contrastive learning loss, is the cross entropy loss, Contribute parameters to adjust the loss, is the contrast loss of each layer; The model parameters of the initial hypergraph convolutional model are optimized by the hierarchical contrastive learning loss until the target hypergraph convolutional model is obtained.

[0050] Specifically, at each level, the contrast loss is calculated by calculating the similarity between drug and disease embeddings. The contrast loss of each layer is:

[0051] , in, is the contrast loss of each layer, and are the drug anchor and disease anchor samples of the lth layer, i.e., the feature data output by each hypergraph convolutional layer. is the temperature parameter, is a distance function, such as cosine similarity.

[0052] The cross entropy loss is: , in, is the cross entropy loss, is the probability predicted by the model, is the weight factor, is a scaling factor used to balance the importance of positive and negative examples during learning.

[0053] Optionally, the interactive information aggregation of the initial drug feature matrix and the initial disease feature matrix to obtain a target drug feature matrix and a target disease feature matrix includes: Obtaining the target drug feature matrix by interactively aggregating information from the initial drug feature matrix; Among them, the target drug feature matrix is: , in, is the target drug feature matrix, is a cross-attention module for improving representation quality, is the initial drug feature matrix; Obtaining the target disease feature matrix by interactively aggregating information from the initial disease feature matrix; Among them, the target disease feature matrix is: , in, is the target disease feature matrix, is the initial disease feature matrix.

[0054] Specifically, in order to better capture the complementary information between different modalities, we introduce the encoder of Transformer and use the self-attention mechanism as the cross-attention module Cross_att(∙) to improve the quality of representation. The low-dimensional representation of drugs and diseases obtained from the last layer of the hypergraph convolutional layer, and as input.

[0055] Optionally, the obtaining of a drug-disease potential association matrix by performing association prediction on the target drug feature matrix and the target disease feature matrix includes: The drug-disease potential association matrix is ​​obtained by performing association prediction on the target drug feature matrix and the target disease feature matrix; Wherein, the drug-disease potential association matrix is: , in, is the drug-disease potential association matrix, is a function that maps predicted values ​​to a preset range, is the target drug feature matrix, is the target disease feature matrix, is the trainable parameter matrix.

[0056] like Figure 2As shown, an embodiment of the present invention provides a drug-disease association prediction device based on a hypergraph, comprising: A hypergraph acquisition module 10, used to acquire a drug hypergraph and a disease hypergraph; An initial feature matrix acquisition module 20 is used to input the drug hypergraph and the disease hypergraph into a target hypergraph convolution model based on a hierarchical contrastive learning algorithm to obtain an initial drug feature matrix and an initial disease feature matrix; A target feature matrix acquisition module 30, configured to perform interactive information aggregation on the initial drug feature matrix and the initial disease feature matrix to obtain a target drug feature matrix and a target disease feature matrix; The drug-disease potential association matrix acquisition module 40 is used to obtain a drug-disease potential association matrix by performing association prediction on the target drug feature matrix and the target disease feature matrix.

[0057] The hypergraph-based drug-disease association prediction device of this embodiment is used to implement the hypergraph-based drug-disease association prediction method as described above. Its advantages over the prior art are the same as the advantages of the hypergraph-based drug-disease association prediction method as described above, and will not be repeated here.

[0058] Optionally, the hypergraph acquisition module 10 is specifically used to: acquire a drug similarity network and a disease similarity network, wherein the drug similarity network includes a drug function similarity network and a drug association similarity network, and the disease similarity network includes a disease semantic similarity hypergraph network and a disease association similarity hypergraph network; A drug hyperedge set and a disease hyperedge set are obtained through the drug similarity network and the disease similarity network, wherein the drug hyperedge set includes a drug function similarity hyperedge group and a drug association similarity hyperedge group, and the disease hyperedge set includes a disease semantic similarity hyperedge group and a disease association similarity hyperedge group; Among them, the drug function similar super marginal group is: , in, is the drug function similarity hyperedge group, k is the number of adjacent drug nodes, is a set of nodes with similar drug functions, v is a vertex of a given drug, is the drug node set; Among them, the drug-associated similarity super-marginal groups are: , in, for the drug-associated similarity super-marginal group, A set of similar nodes associated with drugs; The drug hypergraph and the disease hypergraph are obtained according to the drug hyperedge set and the disease hyperedge set.

[0059] Optionally, the hypergraph acquisition module 10 is specifically used to: obtain a drug adjacency matrix and a disease adjacency matrix respectively through the drug hyperedge set and the disease hyperedge set; Obtaining the drug hypergraph and the disease hypergraph according to the drug adjacency matrix and the disease adjacency matrix respectively; Wherein, the drug hypergraph is: , in, is the drug hypergraph, is the drug adjacency matrix.

[0060] Optionally, the initial feature matrix acquisition module 20 is specifically used to: input the drug hypergraph and the disease hypergraph into the first hypergraph convolution layer of a plurality of sequentially connected hypergraph convolution layers, respectively, and extract features in sequence through the plurality of hypergraph convolution layers to obtain the initial drug feature matrix and the initial disease feature matrix, wherein features are extracted in each of the hypergraph convolution layers by an information propagation algorithm; Wherein, the information propagation algorithm is: , in, is the output feature matrix of the l+1th layer, is any nonlinear activation function, is the diagonal matrix from vertices to hyperedges, H is the adjacency matrix of the hypergraph, and W is the matrix that records the hyperedge weights. is the diagonal matrix from hyperedges to vertices, is the transposed adjacency matrix of the hypergraph, is the preset output feature matrix of the lth layer, is the trainable parameter matrix.

[0061] Optionally, the initial feature matrix acquisition module 20 is specifically used to: obtain an initial hypergraph convolution model; Inputting the training data into the initial hypergraph convolutional model to obtain a prediction result, and calculating the layer-wise contrastive learning loss according to the prediction result; Among them, the layered contrastive learning loss is: , in, is the layer-wise contrastive learning loss, is the cross entropy loss, Contribute parameters to adjust the loss, is the contrast loss of each layer; The model parameters of the initial hypergraph convolutional model are optimized by the hierarchical contrastive learning loss until the target hypergraph convolutional model is obtained.

[0062] Optionally, the target feature matrix acquisition module 30 is specifically used to: obtain the target drug feature matrix by interactively aggregating information from the initial drug feature matrix; Among them, the target drug feature matrix is: , in, is the target drug feature matrix, is a cross-attention module for improving representation quality, is the initial drug feature matrix; Obtaining the target disease feature matrix by interactively aggregating information from the initial disease feature matrix; Among them, the target disease feature matrix is: , in, is the target disease feature matrix, is the initial disease feature matrix.

[0063] Optionally, the drug-disease potential association matrix acquisition module 40 is specifically used to: obtain the drug-disease potential association matrix by performing association prediction on the target drug feature matrix and the target disease feature matrix; Wherein, the drug-disease potential association matrix is: , in, is the drug-disease potential association matrix, is a function that maps predicted values ​​to a preset range, is the target drug feature matrix, is the target disease feature matrix, is the trainable parameter matrix.

[0064] like Figure 3 As shown, an electronic device 300 provided by an embodiment of the present invention includes a memory 310 and a processor 320; the memory 310 is used to store a computer program; the processor 320 is used to implement the hypergraph-based drug-disease association prediction method as described above when executing the computer program.

[0065] In other words, an electronic device 300 includes a memory 310 and a processor 320 coupled to the memory 310; the memory 310 is configured to store a computer program; and the processor 320 is configured to perform the following operations when executing the computer program: Obtain drug hypergraphs and disease hypergraphs; Based on a hierarchical contrastive learning algorithm, the drug hypergraph and the disease hypergraph are input into a target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix; Obtaining a target drug feature matrix and a target disease feature matrix by interactively aggregating the initial drug feature matrix and the initial disease feature matrix; The drug-disease potential association matrix is ​​obtained by performing association prediction on the target drug feature matrix and the target disease feature matrix.

[0066] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the hypergraph-based drug-disease association prediction method as described above is implemented.

[0067] In other words, a non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor performs the following operations: Obtain drug hypergraphs and disease hypergraphs; Based on a hierarchical contrastive learning algorithm, the drug hypergraph and the disease hypergraph are input into a target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix; Obtaining a target drug feature matrix and a target disease feature matrix by interactively aggregating the initial drug feature matrix and the initial disease feature matrix; The drug-disease potential association matrix is ​​obtained by performing association prediction on the target drug feature matrix and the target disease feature matrix.

[0068] An electronic device 300 that can be used as a server or client of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device 300 is intended to represent various forms of digital electronic computer equipment, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 300 can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples, and are not intended to limit the implementation of the present invention described and / or required herein.

[0069] The electronic device 300 includes a computing unit, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The computing unit, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0070] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0071] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A hypergraph-based drug-disease association prediction method, characterized in that: include: Obtain drug hypergraphs and disease hypergraphs; Based on a hierarchical contrastive learning algorithm, the drug hypergraph and the disease hypergraph are input into a target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix; Obtaining a target drug feature matrix and a target disease feature matrix by interactively aggregating the initial drug feature matrix and the initial disease feature matrix; The drug-disease potential association matrix is ​​obtained by performing association prediction on the target drug feature matrix and the target disease feature matrix.

2. The hypergraph-based drug-disease association prediction method according to claim 1, characterized in that: The obtaining of the drug hypergraph and the disease hypergraph comprises: Acquire a drug similarity network and a disease similarity network, wherein the drug similarity network includes a drug function similarity network and a drug association similarity network, and the disease similarity network includes a disease semantic similarity hypergraph network and a disease association similarity hypergraph network; A drug hyperedge set and a disease hyperedge set are obtained through the drug similarity network and the disease similarity network, wherein the drug hyperedge set includes a drug function similarity hyperedge group and a drug association similarity hyperedge group, and the disease hyperedge set includes a disease semantic similarity hyperedge group and a disease association similarity hyperedge group; Among them, the drug function similar super marginal group is: ， in, is the drug function similarity hyperedge group, k is the number of adjacent drug nodes, is a set of nodes with similar drug functions, v is a vertex of a given drug, is the drug node set; Among them, the drug-associated similarity super-marginal groups are: ， in, for the drug-associated similarity supermarginal group, A set of similar nodes associated with drugs; The drug hypergraph and the disease hypergraph are obtained according to the drug hyperedge set and the disease hyperedge set.

3. The hypergraph-based drug-disease association prediction method according to claim 2, characterized in that: The step of obtaining the drug hypergraph and the disease hypergraph according to the drug hyperedge set and the disease hyperedge set includes: Obtaining a drug adjacency matrix and a disease adjacency matrix respectively through the drug hyperedge set and the disease hyperedge set; Obtaining the drug hypergraph and the disease hypergraph according to the drug adjacency matrix and the disease adjacency matrix respectively; Wherein, the drug hypergraph is: ， in, is the drug hypergraph, is the drug adjacency matrix.

4. The hypergraph-based drug-disease association prediction method according to claim 1, characterized in that: The target hypergraph convolution model includes a plurality of hypergraph convolution layers connected in sequence, and the step of inputting the drug hypergraph and the disease hypergraph into the target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix includes: Inputting the drug hypergraph and the disease hypergraph into the first hypergraph convolution layer of a plurality of sequentially connected hypergraph convolution layers respectively, extracting features in sequence through the plurality of hypergraph convolution layers to obtain the initial drug feature matrix and the initial disease feature matrix, wherein features are extracted in each of the hypergraph convolution layers by an information propagation algorithm; Wherein, the information propagation algorithm is: ， in, is the output feature matrix of the l+1th layer, is any nonlinear activation function, is the diagonal matrix from vertices to hyperedges, H is the adjacency matrix of the hypergraph, and W is the matrix that records the hyperedge weights. is the diagonal matrix from hyperedges to vertices, is the transposed adjacency matrix of the hypergraph, is the preset output feature matrix of the lth layer, is the trainable parameter matrix.

5. The hypergraph-based drug-disease association prediction method according to claim 4, characterized in that: The construction process of the target hypergraph convolutional model includes: Get the initial hypergraph convolution model; Inputting the training data into the initial hypergraph convolutional model to obtain a prediction result, and calculating the layer-wise contrastive learning loss according to the prediction result; Among them, the layered contrastive learning loss is: ， in, is the layer-wise contrastive learning loss, is the cross entropy loss, Contribute parameters to adjust the loss, is the contrast loss of each layer; The model parameters of the initial hypergraph convolutional model are optimized by the hierarchical contrastive learning loss until the target hypergraph convolutional model is obtained.

6. The hypergraph-based drug-disease association prediction method according to claim 1, characterized in that: The interactive information aggregation of the initial drug feature matrix and the initial disease feature matrix to obtain a target drug feature matrix and a target disease feature matrix includes: Obtaining the target drug feature matrix by interactively aggregating information from the initial drug feature matrix; Among them, the target drug feature matrix is: ， in, is the target drug feature matrix, is a cross-attention module for improving representation quality, is the initial drug feature matrix; Obtaining the target disease feature matrix by interactively aggregating information from the initial disease feature matrix; Among them, the target disease feature matrix is: ， in, is the target disease feature matrix, is the initial disease feature matrix.

7. The hypergraph-based drug-disease association prediction method according to claim 1, characterized in that: The step of performing association prediction on the target drug feature matrix and the target disease feature matrix to obtain a drug-disease potential association matrix includes: The drug-disease potential association matrix is ​​obtained by performing association prediction on the target drug feature matrix and the target disease feature matrix; Wherein, the drug-disease potential association matrix is: ， in, is the drug-disease potential association matrix, is a function that maps predicted values ​​to a preset range, is the target drug feature matrix, is the target disease feature matrix, is the trainable parameter matrix.

8. A drug-disease association prediction device based on a hypergraph, characterized in that: include: A hypergraph acquisition module, used to acquire drug hypergraphs and disease hypergraphs; An initial feature matrix acquisition module, used for inputting the drug hypergraph and the disease hypergraph into a target hypergraph convolution model to obtain an initial drug feature matrix and an initial disease feature matrix based on a hierarchical contrastive learning algorithm; A target feature matrix acquisition module, used for interactively aggregating information from the initial drug feature matrix and the initial disease feature matrix to obtain a target drug feature matrix and a target disease feature matrix; The drug-disease potential association matrix acquisition module is used to obtain the drug-disease potential association matrix by performing association prediction on the target drug feature matrix and the target disease feature matrix.

9. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the hypergraph-based drug-disease association prediction method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the hypergraph-based drug-disease association prediction method according to any one of claims 1 to 7 is implemented.

Images