Drug-microorganism association prediction method based on multi-view variational graph automatic encoder and multi-scale hypergraph convolution

Abstract

The invention discloses a drug-microorganism association prediction method based on a multi-view variational graph automatic encoder and multi-scale hypergraph convolution, and the method comprises the steps: constructing a first drug-microorganism isomerism graph and a second drug-microorganism isomerism graph, and building local neighbor enhanced attribute embedding; learning attribute distribution of nodes by combining a multi-view variational graph automatic encoder; an enhanced hyperedge attribute is established in combination with cross attention; constructing a dynamically evolved hypergraph topology; establishing a multi-scale hypergraph convolution framework, and adaptively fusing multi-scale node attributes to obtain dynamic hypergraph attribute representation of nodes; through comparative learning of a plurality of graphs, the consistency of features of the same node in the plurality of graphs is enhanced. The prediction method can deduce the correlation tendency between the drugs and the microorganisms obtained in real time. In conclusion, the multi-view node attribute distribution and the multi-scale neighbor topology can be learned, the relation between the multiple nodes of the hypergraph can be dynamically coded, the node representation consistency is enhanced, and the performance of drug-related microorganism candidate prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution. Background Art

[0002] The human microbiome is a complex community of bacteria, archaea, viruses, and microbial eukaryotes. Studies have shown that microbes can enhance or inhibit the efficacy of drugs by changing their structure. The intestinal microbiome affects the immune surveillance of colorectal cancer and the response to immunotherapy. Penicillin can treat pneumonia, meningitis, endocarditis, etc. However, the widespread use of antibiotics has led to drug resistance in human microbes such as Staphylococcus aureus and Escherichia coli. Therefore, predicting drug-microbe associations can screen relevant microbial candidates for a given drug, which helps to explore the mechanism of action of the drug.

[0003] Machine learning methods have been widely used to develop new drug-microbe association prediction methods. For example, the association prediction method based on KATZ measure infers the association between drugs and microorganisms by calculating the path length between them; the Laplace regularized least squares algorithm infers the tendency of drug-microbe association by evaluating the association score between drugs and microorganisms. However, these methods fail to learn the deep features of drugs and microorganisms. With the development of deep learning technology, prediction methods based on deep learning have gradually been introduced into the task of drug-microbe association prediction. For example, graph autoencoders and variational graph autoencoders have been used to learn the feature distribution of nodes, and there are also prediction methods that use drug and microorganism data and combine graph attention networks, meta-path-oriented graph convolution networks, graph neural networks, graph neural networks in hyperbolic space, and relation-sensitive transformers to infer drug-related microbial candidates. However, none of these methods integrate the connections between multiple drug and microbial nodes. To solve this problem, some researchers have tried to integrate such connections based on hypergraph convolution methods, but traditional hypergraph convolution methods usually only focus on the aggregation of node neighbor topology information at a single scale, while ignoring the information of neighbor topologies at multiple scales. In addition, the consistency between representations of the same node in multiple graphs is also ignored, which may limit the predictive performance of the model.

[0004] Therefore, it is necessary to propose a drug-microbe association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution for the drug-microbe association prediction task, which can learn the feature distribution of multi-view drug and microbe nodes, multi-scale neighbor topology and dynamically encode the connections between multiple nodes from the hypergraph, enhance the consistency of node representation, and thus improve the performance of drug-related candidate tasks. Summary of the invention

[0005] In view of this, the present invention provides a drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution, which is used to solve the technical problem that the current drug-microorganism association prediction method does not integrate the relationship between multiple drug and microorganism nodes, and only focuses on the aggregation of node neighbor topology information at a single scale, ignoring the information of neighbor topologies at multiple scales, resulting in low accuracy of association prediction results.

[0006] The present invention provides a drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution, comprising:

[0007] Access to drug-microbe association database;

[0008] According to the drug-microorganism association database, a first heterogeneous graph based on drug function similarity and a second heterogeneous graph based on drug attribute similarity containing drug and microorganism nodes are constructed, local neighbor enhanced attribute embedding is established, and based on the local neighbor enhanced attribute embedding and a multi-view variational graph autoencoder, a heterogeneous graph attribute representation of the node is obtained;

[0009] Generate a drug-microorganism hypergraph according to the drug-microorganism association database, adopt a cross-attention strategy, and construct a dynamic hypergraph topology based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph;

[0010] Introducing a hypergraph Laplacian operator into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, adaptively fusing the neighbor topology embedding to obtain multi-scale node attributes, and supplementing the multi-scale node attributes in the hypergraph convolution layer to obtain a dynamic hypergraph attribute representation of the node;

[0011] According to the heterogeneous graph attribute representation and the dynamic hypergraph attribute representation, graph comparison learning is performed on any two graphs among the first heterogeneous graph, the second heterogeneous graph, and the hypergraph of the multi-scale topological perspective, and a drug-microorganism association prediction model is obtained through training;

[0012] The characteristic vectors of the drug and the microorganism are acquired in real time, and the correlation between the drug and the microorganism is analyzed by the drug-microorganism correlation prediction model.

[0013] Furthermore, the method for establishing attribute embedding for local neighbor enhancement includes:

[0014] In the first heterogeneous graph and the second heterogeneous graph, drugs or microorganisms are used as target nodes, and several neighboring nodes with the most similar attributes to the target nodes are screened; wherein, in the first heterogeneous graph, cosine similarity is used to calculate the similarity of node attributes, and in the second heterogeneous graph, heat kernel similarity is used to calculate the similarity of node attributes;

[0015] Concatenate the attribute vector of the target node with the attribute vector of the neighbor node to obtain an attribute matrix of the neighbor node of the target node;

[0016] The neighbor node attribute matrix is ​​weightedly aggregated using a convolution filter, and the original attribute information is retained through residual connections to obtain attribute embedding for local neighbor enhancement.

[0017] Furthermore, the attribute embedding based on the local neighbor enhancement and the multi-view variational graph autoencoder obtains the heterogeneous graph attribute representation of the node, including:

[0018] Using the attribute embedding enhanced by the local neighbors as an input of a shared parameter graph convolutional coding layer, obtaining common information according to the shared parameter graph convolutional coding layer, and propagating and supplementing the common information to the first heterogeneous graph and the second heterogeneous graph to form an attribute matrix;

[0019] The variational graph autoencoder is used to model the node feature distribution according to the attribute matrix, and the feature distribution of the heterogeneous graph nodes is calculated through the reparameterization technique.

[0020] Furthermore, the attribute embedding enhanced by local neighbors is used as an input of a shared parameter graph convolutional coding layer, common information is obtained according to the shared parameter graph convolutional coding layer, and the common information is propagated and supplemented to the first heterogeneous graph and the second heterogeneous graph to form an attribute matrix, including:

[0021] Encoding the topology and node attributes shared by the first heterogeneous graph and the second heterogeneous graph through a shared parameter graph convolutional network to obtain a shared information feature graph;

[0022] The common information feature graph obtained from the drug function perspective of the first heterogeneous graph is propagated and supplemented to the drug attribute perspective of the second heterogeneous graph to form an attribute matrix of the nodes of the second heterogeneous graph;

[0023] The common information feature graph obtained from the perspective of drug attributes of the second heterogeneous graph is propagated and supplemented to the drug function perspective of the first heterogeneous graph to form an attribute matrix of the nodes of the first heterogeneous graph.

[0024] Furthermore, the variational graph autoencoder is used to model the node feature distribution according to the attribute matrix, and the heterogeneous graph node feature distribution is calculated by reparameterization technique, including:

[0025] Using the first graph convolutional network and the second graph convolutional network, respectively, calculate the mean and variance of the attribute matrix of the first heterogeneous graph and the second heterogeneous graph;

[0026] calculating a latent feature distribution of the node based on a reparameterization technique, wherein the latent feature distribution is sampled from a normal distribution;

[0027] The common information features are concatenated with the potential feature distribution to obtain heterogeneous graph attribute representations of all nodes in the first heterogeneous graph and the second heterogeneous graph.

[0028] Furthermore, the drug-microorganism hypergraph is generated according to the drug-microorganism association database, a cross-attention strategy is adopted, and a dynamic hypergraph topology is constructed based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph, including:

[0029] The hyperedge attribute matrices of the drug-microorganism hypergraph are sampled respectively according to a learnable Gaussian distribution, and are concatenated to form a joint hyperedge attribute matrix; at the same time, the node attribute matrices of the hypergraph are concatenated to form a joint node attribute matrix;

[0030] Aggregating all attributes of nodes in the hypergraph through a cross-attention strategy to generate enhanced hyper-edge attributes;

[0031] A dynamic hypergraph topology is constructed according to the enhanced hyperedge attributes and the joint node attribute matrix.

[0032] Furthermore, all attributes of the nodes in the hypergraph are aggregated through a cross-attention strategy to generate enhanced hyperedge attributes, including:

[0033] Linearly transforming the joint hyperedge attribute matrix to obtain a query matrix;

[0034] Linearly transforming the joint node attribute matrix to obtain a key matrix and a value matrix respectively;

[0035] According to the query matrix, the transpose of the key matrix and the value matrix, enhanced hyperedge attributes are obtained.

[0036] Furthermore, a hypergraph Laplacian operator is introduced into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, and multi-scale node attributes are obtained by adaptive fusion according to the neighbor topology embedding, and the multi-scale node attributes are supplemented in the hypergraph convolution layer to obtain a dynamic hypergraph attribute representation of the node, including:

[0037] Normalizing the dynamic hypergraph topology according to the hypergraph Laplacian operator to form a multi-order neighbor topology embedding;

[0038] Generate a topological attribute matrix of multi-order neighbor nodes by using the multi-order neighbor topology embedding;

[0039] Adaptively fuse the topological attribute matrices of all scales to form a multi-scale node attribute matrix from a multi-scale topological perspective;

[0040] For any hypergraph convolutional layer in the hypergraph, the topological attribute matrix of the current convolutional layer and the multi-scale node attribute representation of the previous convolutional layer are concatenated until all hypergraph convolutional layers in the hypergraph are concatenated to obtain the dynamic hypergraph attribute representation of all nodes.

[0041] Furthermore, according to the heterogeneous graph attribute representation and the dynamic hypergraph attribute representation, graph comparison learning is performed on any two graphs in the first heterogeneous graph, the second heterogeneous graph, and the hypergraph of the multi-scale topological perspective, and a drug-microorganism association prediction model is trained, including:

[0042] A first loss function of the first heterogeneous graph and the dynamic hypergraph, and a second loss function of the second heterogeneous graph and the dynamic hypergraph are calculated according to the heterogeneous graph attribute representation of the node and the dynamic hypergraph attribute representation of the node;

[0043] A third loss function is calculated based on the heterogeneous graph attribute representation of the nodes in the first heterogeneous graph and the attribute representation of the nodes in the second heterogeneous graph;

[0044] A joint contrast loss function is calculated according to the first loss function, the second loss function and the third loss function, and a drug-microorganism association prediction model is obtained according to the joint contrast loss function.

[0045] The present invention also provides a drug-microorganism association prediction system based on multi-view variational graph autoencoder and multi-scale hypergraph convolution, comprising:

[0046] A data acquisition module, used to acquire a drug-microorganism association database;

[0047] A multi-view variational graph autoencoder module is used to construct a first heterogeneous graph based on drug function similarity and a second heterogeneous graph based on drug attribute similarity containing drug and microorganism nodes according to the drug-microorganism association database, establish a local neighbor enhanced attribute embedding, and obtain a heterogeneous graph attribute representation of the node based on the local neighbor enhanced attribute embedding and the multi-view variational graph autoencoder;

[0048] A dynamic hypergraph module, used for generating a drug-microorganism hypergraph according to the drug-microorganism association database, adopting a cross-attention strategy, and constructing a dynamic hypergraph topology based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph;

[0049] A multi-scale convolution module, used to introduce a hypergraph Laplacian operator into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, adaptively fuse the neighbor topology embedding to obtain multi-scale node attributes, and supplement the multi-scale node attributes in the hypergraph convolution layer to obtain a dynamic hypergraph attribute representation of the node;

[0050] A comparative learning module, used for performing graph comparative learning on any two graphs among the first heterogeneous graph, the second heterogeneous graph and the hypergraph of multi-scale topological perspective according to the heterogeneous graph attribute representation and the dynamic hypergraph attribute representation, and training to obtain a drug-microorganism association prediction model;

[0051] The association prediction module is used to obtain the characteristic vectors of drugs and microorganisms in real time, and analyze the association between the drugs and microorganisms through the drug-microorganism association prediction model.

[0052] Compared with the prior art, the advantages of the present invention are:

[0053] (1) We construct drug-microbe heterogeneous graphs that contain unique information from the drug function perspective and the drug attribute perspective, and design a variational graph autoencoder that propagates this unique information in both perspectives to model the feature distribution of each drug and microbe node. This helps reveal the feature distribution of each node in the entire graph. Unlike traditional techniques that only model a single perspective, this method can flexibly adapt to a variety of different datasets and problem scenarios, rather than being limited to a single or specific scenario.

[0054] (2) A hypergraph composed of hyperedges is used to reveal the close connections between multiple drugs and microorganisms from multiple biological perspectives. This method introduces hyperedge attributes for dynamic learning, which can adaptively capture the association between drug and microbial nodes in different biological contexts and form a topological structure of a dynamically evolving hypergraph, which is scalable.

[0055] (3) A multi-scale topological hypergraph convolution framework is used to generate a drug-microbe node neighbor topology embedding covering multiple scales, and adaptively integrate neighbor information at different scales. This method solves the defect that traditional hypergraph convolution only targets a single neighbor topology.

[0056] (4) A multiple graph comparison learning strategy is adopted to maximize the similarity of node feature representations in heterogeneous graphs and hypergraphs, effectively maintaining the consistency of node representations from multiple graph perspectives, which improves the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of the process of the drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution provided by the present invention;

[0058] Figure 2 A graphical data flow diagram provided by the present invention;

[0059] Figure 3 A schematic diagram of data processing for the multi-view variational graph autoencoder provided by the present invention;

[0060] Figure 4 A schematic diagram of dynamic hyperedge attribute learning provided by the present invention;

[0061] Figure 5 A schematic diagram of the process of multi-scale hypergraph convolution provided by the present invention;

[0062] Figure 6 A schematic diagram of the multi-image comparison process provided by the present invention. DETAILED DESCRIPTION

[0063] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0064] The present invention provides a drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution, which is described in detail below.

[0065] See also Figure 1 This embodiment provides a drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution, including:

[0066] Step S101: Acquire a drug-microorganism association database;

[0067] Step S102: constructing a first heterogeneous graph based on drug function similarity and a second heterogeneous graph based on drug attribute similarity containing drug and microorganism nodes according to the drug-microorganism association database, establishing a local neighbor enhanced attribute embedding, and obtaining a heterogeneous graph attribute representation of the node based on the local neighbor enhanced attribute embedding and a multi-view variational graph autoencoder;

[0068] Step S103: generating a drug-microorganism hypergraph according to the drug-microorganism association database, and using a cross-attention strategy to construct a dynamic hypergraph topology based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph;

[0069] Step S104: introducing a hypergraph Laplacian operator into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, adaptively fusing node attributes from a multi-scale topology perspective according to the neighbor topology embedding, and constructing a hypergraph from a multi-scale topology perspective to obtain a hypergraph attribute representation of the node;

[0070] Step S105: performing graph comparison learning on any two graphs among the first heterogeneous graph, the second heterogeneous graph and the hypergraph from a multi-scale topological perspective according to the heterogeneous graph attribute representation and the hypergraph attribute representation, and training to obtain a drug-microorganism association prediction model;

[0071] Step S106: acquiring the characteristic vectors of the drug and the microorganism in real time, and analyzing the correlation between the drug and the microorganism by using the drug-microorganism correlation prediction model.

[0072] The method of this embodiment learns multi-perspective node feature distribution in multiple heterogeneous graphs and hypergraphs, dynamically encodes the connections between multiple nodes, enhances the consistency of node representation, and helps to screen potential microbial candidates for drugs.

[0073] As a specific embodiment, in step S101, the drug-microorganism association database includes: associations between drugs and microorganisms, similarity of drug chemical substructures, drug Gaussian kernel similarity and microorganism similarity; wherein, the associations between drugs and microorganisms include 2470 pairs of associations between drugs and microorganisms supported by clinical or experimental methods originally from the MDAD database (Sun et al., 2018), which contain 1373 drugs and 173 microorganisms. Drug chemical substructure similarity is calculated based on the maximum common subgraph of drug structure by SIMCOMP2 (Hattori et al., 2010), from the previous work of EGATMDA (Long et al., 2020a). Drug Gaussian kernel similarity is based on the drugs that interact with two drugs to calculate the Gaussian kernel similarity between them. The similarity of microorganisms is calculated by the cosine similarity of microorganism sequence features, and the gene sequences of microorganisms are obtained from the National Center for Biotechnology (NCBI) and key sequence features are extracted by principal component analysis] (PCA) (Chen et al., 2002).

[0074] As a specific embodiment, in step S102, the first heterogeneous graph is constructed based on the functional perspective of drug-drug. The nodes in the graph include drug and microorganism nodes, and the edges include drug-drug edges and drug-microorganism edges. Drug-drug edges represent functional similarities between drugs, and these edges can be based on information such as shared targets, mechanisms of action, therapeutic effects or pharmacological categories of drugs; drug-microorganism edges represent the effects of drugs on microorganisms, such as inhibition, promotion or no effect, and these edges can be based on experimental data and clinical research information.

[0075] The second heterogeneous graph is constructed based on the drug-drug attribute perspective. The nodes in the graph include drug and microorganism nodes. In the edge relationship, the drug-drug edge represents the attribute similarity between drugs. These edges can be based on the drug's chemical structure, physicochemical properties (such as solubility, stability), molecular function and other information. The drug-microorganism edge represents the effect of the drug on the microorganism, such as inhibition, promotion or no effect.

[0076] In order to learn the node feature distribution from multiple perspectives, it is necessary to determine the common information first. The present invention uses the neighbor-enhanced attribute embedding as the input of the shared parameter graph convolutional coding layer. For a preferred embodiment, please refer to Figure 2 In part (a), the establishment method of attribute embedding for local neighbor enhancement includes:

[0077] In the first heterogeneous graph and the second heterogeneous graph, respectively, drugs or microorganisms are used as target nodes, and top-k neighbor nodes with the most similar attributes to the target nodes are screened; wherein, in the first heterogeneous graph, cosine similarity is used to calculate the similarity of node attributes, and in the second heterogeneous graph, heat kernel similarity is used to calculate the similarity of node attributes;

[0078] Concatenate the attribute vector of the target node with the attribute vector of the neighbor node to obtain an attribute matrix of the neighbor node of the target node;

[0079] A 1×1 convolution filter is used to perform weighted aggregation on the neighbor node attribute matrix, and the original attribute information is retained through residual connections to obtain attribute embedding for local neighbor enhancement.

[0080] By establishing local neighbor-enhanced attribute embedding through the above method, the target node can obtain more attribute information from different neighbors, which helps to establish a richer multi-level feature representation. In the task of drug and microbial association prediction, the attributes of the node are not only determined by a single feature, but are obtained by interacting with the features of multiple adjacent nodes. The enhanced embedding can better capture this complex feature interaction, thereby mining the key information hidden in the complex interactions between nodes.

[0081] As a preferred embodiment, in step S102, based on the local neighbor enhanced attribute embedding and the multi-view variational graph autoencoder, a heterogeneous graph attribute representation of a node is obtained, including:

[0082] Using the attribute embedding enhanced by the local neighbors as an input of a shared parameter graph convolutional coding layer, obtaining common information according to the shared parameter graph convolutional coding layer, and propagating and supplementing the common information to the first heterogeneous graph and the second heterogeneous graph to form an attribute matrix;

[0083] The variational graph autoencoder is used to model the node feature distribution according to the attribute matrix, and the feature distribution of the heterogeneous graph nodes is calculated through the reparameterization technique.

[0084] As a preferred embodiment, the attribute embedding enhanced by local neighbors is used as the input of the shared parameter graph convolutional coding layer, common information is obtained according to the shared parameter graph convolutional coding layer, and the common information is propagated and supplemented to the first heterogeneous graph and the second heterogeneous graph to form an attribute matrix, including:

[0085] Encoding the topology and node attributes shared by the first heterogeneous graph and the second heterogeneous graph through a shared parameter graph convolutional network to obtain a shared information feature graph;

[0086] The common information feature graph obtained from the drug function perspective of the first heterogeneous graph is propagated and supplemented to the drug attribute perspective of the second heterogeneous graph to form an attribute matrix of the nodes of the second heterogeneous graph;

[0087] The common information feature graph obtained from the perspective of drug attributes of the second heterogeneous graph is propagated and supplemented to the drug function perspective of the first heterogeneous graph to form an attribute matrix of the nodes of the first heterogeneous graph.

[0088] As a preferred embodiment, the method of using a variational graph autoencoder to model the node feature distribution according to the attribute matrix and calculating the heterogeneous graph attribute representation of the node by a reparameterization technique includes:

[0089] Using the first graph convolutional network and the second graph convolutional network, respectively, calculate the mean and variance of the attribute matrix of the first heterogeneous graph and the second heterogeneous graph;

[0090] calculating a latent feature distribution of the node based on a reparameterization technique, wherein the latent feature distribution is sampled from a normal distribution;

[0091] The common information features are concatenated with the potential feature distribution to obtain heterogeneous graph attribute representations of all nodes in the first heterogeneous graph and the second heterogeneous graph.

[0092] Combine the following Figure 2 The process of learning the feature distribution of nodes using the multi-view variational graph autoencoder method is described in detail using a specific embodiment. Figure 2 Part (b) of

[0093] Step S201: Setting model parameters of the first isomorphic graph and the second isomorphic graph, including the drug-microorganism isomorphic graph The adjacency matrix of Node property matrix Attribute embedding for local neighborhood enhancement Independent First Graph Convolutional Network And the second graph convolutional network And parameter-sharing graph convolutional network GCN co ;in, fu and he represent two different types of drug-drug similarities, respectively.

[0094] Step S202: Initialize parameters, including the weight matrices of the independent first and second graph convolutional networks Weight matrix W of parameter-sharing graph convolutional network co .

[0095] Step S203: embedding the local neighbor enhanced attributes into Y fu ,Y he As parameter sharing graph convolution GCN co The input of the drug and microbial nodes is obtained to obtain the common information feature graph C fu ,C he .

[0096] Step S204: The feature map C he and the node attribute matrix F fu Spread and replenish G fu Perspective, forming an attribute matrix

[0097] Step S205: The feature map C fu and the node attribute matrix F he Spread and replenish G he Perspective, forming an attribute matrix

[0098] Step S206: Using graph convolution and calculate The mean and variance And based on the reparameterization technology, the characteristic distribution of nodes is calculated in, From the normal distribution Medium sampling.

[0099] Step S207: stitching feature maps And the characteristic distribution of nodes Get all node attribute representations

[0100] like Figure 3 As shown, Figure 3The data processing flow chart of the multi-view variational graph autoencoder is shown. Through the above steps, the shared graph convolutional network is used to fuse the information in the two heterogeneous graphs, so that each graph can obtain complementary information from the other graph, which enhances the model's understanding of node features. In addition, by combining the output of the graph convolutional network with the variational autoencoder, the graph convolutional network is used to extract the features of the nodes, and the node features are mapped to the mean and variance of the latent space through the fully connected layer, and the parameters of the latent space distribution are learned (similar to the encoder in VAE, except that the node features output from GCN are used to generate the distribution parameters of the latent space). This method can learn a low-dimensional representation of the node to capture the potential features of the node, which is helpful for subsequent tasks such as node classification and link prediction. By propagating information from one perspective to another, it can supplement the lack of information from different perspectives and improve the model's ability to understand complex relationships. The reparameterization technique enables the model to be optimized by stochastic gradient descent, which improves the efficiency of training.

[0101] In a specific embodiment, in the above steps S204 and S205, the feature map C he (C fu ) and the node attribute matrix F fu (F he ) spread and replenish G fu (G he ) perspective, and is expressed as follows:

[0102]

[0103] Among them, P he ,P fu is the learnable parameter matrix, W sf ,W sh is the weight matrix.

[0104] It can be seen that the parameter matrix P he ,P fu It is learnable, and the model can automatically adjust these parameters to optimize performance. By combining features and attributes from different heterogeneous graphs, the expressiveness of node features can be enhanced, allowing the model to capture more complex patterns and relationships. By propagating attributes across graphs, the model can better generalize to unseen data and is suitable for complex tasks that require the integration of multiple information to complete.

[0105] As a preferred embodiment, in step S103, a drug-microorganism hypergraph is generated according to the drug-microorganism association database, and a cross-attention strategy is adopted to construct a dynamic hypergraph topology based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph, including:

[0106] The hyperedge attribute matrices of the drug-microorganism hypergraph are sampled respectively according to a learnable Gaussian distribution, and are concatenated to form a joint hyperedge attribute matrix; at the same time, the node attribute matrices of the hypergraph are concatenated to form a joint node attribute matrix;

[0107] Aggregating all attributes of nodes in the hypergraph through a cross-attention strategy to generate enhanced hyper-edge attributes;

[0108] A dynamic hypergraph topology is constructed according to the enhanced hyperedge attributes and the joint node attribute matrix.

[0109] As a preferred embodiment, all attributes of nodes in the hypergraph are aggregated through a cross-attention strategy to generate enhanced hyperedge attributes, including:

[0110] Linearly transforming the joint hyperedge attribute matrix to obtain a query matrix;

[0111] Linearly transforming the joint node attribute matrix to obtain a key matrix and a value matrix respectively;

[0112] According to the query matrix, the transpose of the key matrix and the value matrix, enhanced hyperedge attributes are obtained.

[0113] As a specific embodiment, Figure 2 Part (c) of Figure 3 illustrates the above process. Hyperedge attribute learning includes the following steps:

[0114] Step S301: Input the joint hyperedge attribute matrix Combined node attribute matrix F mn ;

[0115] Step S302: Initialize parameters;

[0116] Step S303: Combine the hyperedge attribute matrix Linear transformation obtains the query matrix Q;

[0117] Step S304: Combine the node attribute matrix F mn Linear transformations are performed to obtain the key matrix K and the value matrix V respectively;

[0118] Step S305: According to Q, K T and V compute the enhanced hyperedge properties

[0119] Step S306: Based on the enhanced hyperedge attributes and the joint node attribute matrix F mn The product of the transposes of forms a dynamic hypergraph topology.

[0120] like Figure 4 As shown, Figure 4The schematic diagram of dynamic hyperedge attribute learning is shown. Through the above steps, the attribute matrices of all hyperedges are spliced ​​together to form a joint hyperedge attribute matrix. This helps to uniformly process the attribute information of all hyperedges in the model, thereby improving the efficiency and performance of the model. Similarly, the attribute matrices of the nodes can also be spliced ​​to form a joint node attribute matrix, making the model more focused and efficient when processing node information. The cross-attention mechanism allows the model to interact with information between different inputs, which can enhance the model's ability to aggregate node attributes, thereby improving the expressiveness of hyperedge attributes. Based on the enhanced hyperedge attributes and the joint node attribute matrix, a dynamic hypergraph topology can be constructed, which can be dynamically adjusted according to the attributes and relationships of the data, thereby better reflecting the intrinsic structure and relationships of the data.

[0121] As a specific embodiment, in step S305, according to Q, K T and V compute the enhanced hyperedge properties The specific method is as follows:

[0122]

[0123] Where N is the total number of nodes.

[0124] Since two nodes with similar attribute distributions are more likely to belong to the same hyperedge, the attribute distributions of each node are combined to learn the hyperedge attributes.

[0125] As a preferred embodiment, in step S104, a hypergraph Laplacian operator is introduced into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, a multi-scale node attribute matrix is ​​obtained by adaptive fusion according to the neighbor topology embedding, and the multi-scale node attribute matrix is ​​supplemented in the hypergraph convolution layer to obtain a dynamic hypergraph attribute representation of the node, including:

[0126] Normalizing the dynamic hypergraph topology according to the hypergraph Laplacian operator to form a multi-order neighbor topology embedding;

[0127] Generate a topological attribute matrix of multi-order neighbor nodes by using the multi-order neighbor topology embedding;

[0128] Adaptively fuse the topological attribute matrices of all scales to form a multi-scale node attribute matrix from a multi-scale topological perspective;

[0129] For any hypergraph convolutional layer in the hypergraph, the topological attribute matrix of the current convolutional layer and the multi-scale node attribute representation of the previous convolutional layer are concatenated until all hypergraph convolutional layers in the hypergraph are concatenated to obtain the dynamic hypergraph attribute representation of all nodes.

[0130] The specific processing process of the above method is described below through a specific embodiment.

[0131] In actual processing, see Figure 2 Part (d) of the process consists of the following steps:

[0132] Step S401: Input enhanced hyperedge attributes Combined node attribute matrix F mn , dynamic hypergraph topology H dy and the hypergraph Laplacian θ;

[0133] Step S402: Initialize parameters, including the number of hypergraph convolution layers T and the order s of neighbor nodes;

[0134] Step S403: According to the hypergraph Laplacian operator θ, the input dynamic hypergraph topology H dy Normalize and form an s-order neighbor topology embedded in A s ;

[0135] Step S404: In the first hypergraph convolutional layer, the attributes of the hyperedges are aggregated based on the hypergraph topology to generate the attribute matrix of the first-order neighbor nodes.

[0136] Step S405: Embed A using s-order neighbor topology s Generate the attribute matrix of s-order neighbor nodes as

[0137] Step S406: Adaptively fuse neighbor topological attribute matrices of all scales to form a node attribute matrix K from a multi-scale topological perspective ms ;

[0138] Step S407: In the lth hypergraph convolutional layer, concatenate the node attribute matrix K ms and the node attribute representation obtained by the l-1th graph convolutional layer;

[0139] Step S408: After T cycles, the loop ends and the node attribute representation K is obtained. jo .

[0140] like Figure 5 As shown, Figure 5 The schematic diagram of the multi-scale hypergraph convolution process is shown. The above method uses a multi-scale topological hypergraph convolution framework to generate drug-microbe node neighbor topological embedding covering multiple scales, and adaptively integrates neighbor information of different scales, solving the defect that traditional hypergraph convolution only targets a single neighbor topology, and can encode the connection between multiple drug and microbe nodes.

[0141] Furthermore, in step S405, a hypergraph Laplacian operator is introduced to form an s-order neighbor topology embedding, which is generated as follows:

[0142]

[0143] Among them, W e is the weight matrix, and the node degree matrix (D v ) ii =∑ j (H dy ) ij , the hyper-edge degree matrix (D e ) jj =∑ i (H dy ) ij .

[0144] By generating s-order neighbor topological embeddings, it helps to form topological embeddings for each drug and microorganism node in the hypergraph, and these embeddings cover the neighbor regions with multiple ranges.

[0145] As a preferred embodiment, in step S105, graph comparison learning is performed on any two graphs in the first heterogeneous graph, the second heterogeneous graph, and the hypergraph of the multi-scale topological perspective according to the heterogeneous graph attribute representation and the dynamic hypergraph attribute representation, and a drug-microorganism association prediction model is trained, including:

[0146] A first loss function of the first heterogeneous graph and the dynamic hypergraph, and a second loss function of the second heterogeneous graph and the dynamic hypergraph are calculated according to the heterogeneous graph attribute representation of the node and the dynamic hypergraph attribute representation of the node;

[0147] A third loss function is calculated based on the heterogeneous graph attribute representation of the nodes in the first heterogeneous graph and the attribute representation of the nodes in the second heterogeneous graph;

[0148] A joint contrast loss function is calculated according to the first loss function, the second loss function and the third loss function, and a drug-microorganism association prediction model is obtained according to the joint contrast loss function.

[0149] By comparing multiple graphs, the model can comprehensively consider different graph structures and attribute representations and enhance the consistency of node representation, thereby more effectively integrating feature information from different graph structures and more accurately predicting the association between drugs and microorganisms.

[0150] As a specific embodiment, the specific processing steps of multiple image comparison include:

[0151] Step S501: Input the heterogeneous graph G fu ,G he Learn to get the node attribute representation K fu and K he , from the hypergraph The learned node attribute matrix K jo, the loss function L fc ,L hc ,L fh ;

[0152] Step S502: Initialize parameters, including temperature parameter τ;

[0153] Step S503: For the heterogeneous graph G fu (G he ) and hypergraph Perform graph contrast learning: Represent K according to node attributes fu (K he ) and the node attribute matrix K jo Calculate the loss function L fc (L hc );

[0154] Step S504: For the heterogeneous graph G fu and heterogeneous graph G he Perform graph contrast learning: Represent K according to node attributes fu and K he Calculate the loss function L fh ;

[0155] Step S505: Calculate the joint contrast loss function.

[0156] like Figure 6 As shown, Figure 6 The schematic diagram of the multi-graph comparison process is shown. Through multiple graph comparison learning, the consistency of node representation of heterogeneous graphs and hypergraphs under different graph perspectives can be maintained, and the model's prediction ability for drug-microorganism association can be effectively improved.

[0157] As a specific example, in order to improve the model's ability to predict drug-microorganism associations and maintain consistency in node representation between heterogeneous graphs and hypergraphs from different graph perspectives, the loss function L fc ,L hc ,L fh The calculation method is as follows:

[0158]

[0159] Among them, sim(·) is the cosine similarity and τ is the temperature parameter.

[0160] It should be noted that in step S106, when the drug-microorganism association prediction model provided by the present invention is used to calculate the association between the drug and the microorganism, since it is oriented to the heterogeneous graph G fu , G he and Hypergraph The corresponding learned attribute matrix of all drugs and microorganisms is K β The i-th row and the N-th rowd +j rows are drug nodes d i and microbial node m j The attribute vector and d i and m j The attribute embedding B β (β∈{fu,he,jo}) is formed,

[0161]

[0162] B fu , B he and B jo Stacking at the channel level

[0163] B=[B fu ; B he ; B jo ],

[0164] In which, “;” indicates a stacking operation. In order to preserve the marginal information of B, we perform zero padding in each convolutional layer, denoted as After two convolution and pooling layers, we get d i -m j The attribute represents J. J is flattened into a vector representing J f And is fed to the fully connected layer to get d i -m j Relevance score,

[0165] η=softmax(W f J f +b f ),

[0166] Among them, W f and b f are the weight matrix and bias vector respectively. η=[(η)0,(η)1], (η)0 and (η)1 represent drug d i and microorganisms j Probability of being uncorrelated or correlated.

[0167] The cross entropy between the true labels of drug-microbe node pairs and their predicted association scores is calculated as the loss function of the model, namely:

[0168]

[0169] Among them, N t is the number of training sample sets and C = 2. If there is a real association between drugs and microorganisms, y j =1, otherwise 0.

[0170] In order to verify the actual effect of the present invention, the prediction performance of this method and other comparative methods was evaluated by five-fold cross validation. The positive samples consisted of all known associations between drugs and microorganisms, and they were randomly divided into 5 subsets of equal size. All drug-microorganism node pairs that have not been found to have an association were regarded as negative samples. In each fold, we selected 4 positive sample subsets and an equal number of randomly selected negative samples as training sets, and the remaining positive sample subsets and all remaining negative samples were used to test the model.

[0171] Evaluation indicators include the receiver operating characteristic curve (ROC) area (AUC), the area under the precision-recall (PR) curve (AUPR), and the recall rate of the top-k drug-microbe candidates. We calculated the AUC and AUPR of each fold during cross-validation and took their average as the final prediction performance of the model.

[0172] The other eight models used for comparison are: DHDMP, MFLP, NGMDA, SCSMDA, GACNNMDA, GSAMDA, EGATMDA and GCNMDA. Among them, the number of encoding layers P in the hypergraph convolution is 2, and the order of neighbor nodes s = 2. The temperature parameter τ in multiple graph comparison learning is 0.1. The AUC and AUPR of this method and the other eight models for drug-microorganism association prediction are shown in Table 1.

[0173] Table 1 Comparison of association prediction results of each model

[0174]

[0175] From Table 1, we can find that the method has the highest average AUC of 96.3%, which is 0.4% higher than DHDMP, 1.7% higher than MFLP, 1.9% higher than NGMDA, 4.7% higher than SCSMDA, 12.0% higher than GACNNMDA, 6.1% higher than GSAMDA, 2.3% higher than EGATMDA, and 6.0% higher than GCNMDA. The average AUPR of MMDMA is 88.0%, which is 5.7%, 11.3%, 15.2%, 54.0%, 68.4%, 63.3%, 57.3% and 56.5% higher than DHDMP, MFLP, NGMDA, SCSMDA, CACNNMDA, GSAMDA, EGATMDA and GCNMDA, respectively.

[0176] The poor performance of GACNNMDA and GSAMDA may be due to the fact that they only learn node features based on GAT without considering the feature distribution of nodes in heterogeneous graphs. GCNMDA and EGATMDA aggregate the neighbor node information of the target node based on graph convolution and node-level attention, respectively, and both achieve good performance. This shows that aggregating neighbor node information is necessary. SCSMDA only focuses on learning meta-path node features and ignores the learning of node features for the entire heterogeneous graph. NGMDA achieves better prediction performance than SCSMDA, which learns the features of drug and microbial nodes from the neighborhood perspective and the entire heterogeneous graph perspective, respectively. MFLP mainly focuses on learning the hierarchical features of multi-class nodes in hyperbolic space. However, these methods ignore the close connections between multiple drug and microbial nodes. DHDMP predicts the association between drugs and microbes based on class-sensitive hypergraph convolution, but ignores the information propagation from nodes to hyperedges. The main reason why our method achieves the best performance is that it learns the connections between multiple nodes and the distribution of node features.

[0177] The embodiment of the present invention also provides a drug-microorganism association prediction system based on multi-view variational graph autoencoder and multi-scale hypergraph convolution, comprising:

[0178] A data acquisition module, used to acquire a drug-microorganism association database;

[0179] A multi-view variational graph autoencoder module is used to construct a first heterogeneous graph based on drug function similarity and a second heterogeneous graph based on drug attribute similarity containing drug and microorganism nodes according to the drug-microorganism association database, establish a local neighbor enhanced attribute embedding, and obtain a heterogeneous graph attribute representation of the node based on the local neighbor enhanced attribute embedding and the multi-view variational graph autoencoder;

[0180] A dynamic hypergraph module, used for generating a drug-microorganism hypergraph according to the drug-microorganism association database, adopting a cross-attention strategy, and constructing a dynamic hypergraph topology based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph;

[0181] A multi-scale convolution module, used to introduce a hypergraph Laplacian operator into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, adaptively fuse the neighbor topology embedding to obtain multi-scale node attributes, and supplement the multi-scale node attributes in the hypergraph convolution layer to obtain a dynamic hypergraph attribute representation of the node;

[0182] A comparative learning module, used for performing graph comparative learning on any two graphs among the first heterogeneous graph, the second heterogeneous graph and the hypergraph of multi-scale topological perspective according to the heterogeneous graph attribute representation and the dynamic hypergraph attribute representation, and training to obtain a drug-microorganism association prediction model;

[0183] The association prediction module is used to obtain the characteristic vectors of drugs and microorganisms in real time, and analyze the association between the drugs and microorganisms through the drug-microorganism association prediction model.

[0184] In summary, the drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution provided in this embodiment can learn the feature distribution of multi-view drug and microorganism nodes, multi-scale neighbor topology and dynamically encode the connections between multiple nodes from the hypergraph, enhance the consistency of node representation, and thus improve the performance of drug-related candidate tasks.

[0185] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A drug-microbe association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution, characterized in that: include: Access to drug-microbe association database; According to the drug-microorganism association database, a first heterogeneous graph based on drug function similarity and a second heterogeneous graph based on drug attribute similarity containing drug and microorganism nodes are constructed, local neighbor enhanced attribute embedding is established, and based on the local neighbor enhanced attribute embedding and a multi-view variational graph autoencoder, a heterogeneous graph attribute representation of the node is obtained; Generate a drug-microorganism hypergraph according to the drug-microorganism association database, adopt a cross-attention strategy, and construct a dynamic hypergraph topology based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph; Introducing a hypergraph Laplacian operator into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, adaptively fusing the neighbor topology embedding to obtain multi-scale node attributes, and supplementing the multi-scale node attributes in the hypergraph convolution layer to obtain a dynamic hypergraph attribute representation of the node; According to the heterogeneous graph attribute representation and the dynamic hypergraph attribute representation of the nodes, graph comparison learning is performed on any two graphs among the first heterogeneous graph, the second heterogeneous graph, and the hypergraph of the multi-scale topological perspective, and a drug-microorganism association prediction model is obtained through training; The characteristic vectors of the drug and the microorganism are acquired in real time, and the correlation between the drug and the microorganism is analyzed by the drug-microorganism correlation prediction model.

2. The drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution according to claim 1 is characterized in that: The method for establishing local neighbor enhanced attribute embedding comprises: In the first heterogeneous graph and the second heterogeneous graph, drugs or microorganisms are used as target nodes, and several neighboring nodes with the most similar attributes to the target nodes are screened; wherein, in the first heterogeneous graph, cosine similarity is used to calculate the similarity of node attributes, and in the second heterogeneous graph, heat kernel similarity is used to calculate the similarity of node attributes; Concatenate the attribute vector of the target node with the attribute vector of the neighbor node to obtain an attribute matrix of the neighbor node of the target node; The neighbor node attribute matrix is ​​weightedly aggregated using a convolution filter, and the original attribute information is retained through residual connections to obtain attribute embedding for local neighbor enhancement.

3. The drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution according to claim 1 is characterized in that: The multi-view variational graph autoencoder obtains a heterogeneous graph attribute representation of a node, including: Using the attribute embedding enhanced by the local neighbors as an input of a shared parameter graph convolutional coding layer, obtaining common information according to the shared parameter graph convolutional coding layer, and propagating and supplementing the common information to the first heterogeneous graph and the second heterogeneous graph to form an attribute matrix; The variational graph autoencoder is used to model the node feature distribution according to the attribute matrix, and the node feature distribution of the heterogeneous graph is calculated through the reparameterization technique.

4. The drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution according to claim 3 is characterized in that: The attribute embedding enhanced by local neighbors is used as an input of a shared parameter graph convolutional coding layer, common information is obtained according to the shared parameter graph convolutional coding layer, and the common information is propagated and supplemented to the first heterogeneous graph and the second heterogeneous graph to form an attribute matrix, including: Encoding the topology and node attributes shared by the first heterogeneous graph and the second heterogeneous graph through a shared parameter graph convolutional network to obtain a shared information feature graph; The common information feature graph obtained from the drug function perspective of the first heterogeneous graph is propagated and supplemented to the drug attribute perspective of the second heterogeneous graph to form an attribute matrix of the nodes of the second heterogeneous graph; The common information feature graph obtained from the perspective of drug attributes of the second heterogeneous graph is propagated and supplemented to the drug function perspective of the first heterogeneous graph to form an attribute matrix of the nodes of the first heterogeneous graph.

5. The drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution according to claim 4 is characterized in that: The method of using the variational graph autoencoder to model the node feature distribution according to the attribute matrix and calculating the heterogeneous graph node feature distribution by reparameterization techniques includes: Using the first graph convolutional network and the second graph convolutional network, respectively, calculate the mean and variance of the attribute matrix of the first heterogeneous graph and the second heterogeneous graph; calculating a latent feature distribution of the node based on a reparameterization technique, wherein the latent feature distribution is sampled from a normal distribution; The common information features are concatenated with the potential feature distribution to obtain heterogeneous graph attribute representations of all nodes in the first heterogeneous graph and the second heterogeneous graph.

6. The drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution according to claim 1, characterized in that: The method of generating a drug-microorganism hypergraph according to the drug-microorganism association database, adopting a cross-attention strategy, and constructing a dynamic hypergraph topology based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph includes: The hyperedge attribute matrices of the drug-microorganism hypergraph are sampled respectively according to a learnable Gaussian distribution, and are concatenated to form a joint hyperedge attribute matrix; at the same time, the node attribute matrices of the hypergraph are concatenated to form a joint node attribute matrix; Aggregating all attributes of nodes in the hypergraph through a cross-attention strategy to generate enhanced hyper-edge attributes; A dynamic hypergraph topology is constructed according to the enhanced hyperedge attributes and the joint node attribute matrix.

7. The drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution according to claim 6, characterized in that: All attributes of the nodes in the hypergraph are aggregated through the cross-attention strategy to generate enhanced hyper-edge attributes, including: Linearly transforming the joint hyperedge attribute matrix to obtain a query matrix; Linearly transforming the joint node attribute matrix to obtain a key matrix and a value matrix respectively; According to the query matrix, the transpose of the key matrix and the value matrix, enhanced hyperedge attributes are obtained.

8. The drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution according to claim 1, characterized in that: Introducing a hypergraph Laplacian operator into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, adaptively fusing the neighbor topology embedding to obtain multi-scale node attributes, and supplementing the multi-scale node attributes in the hypergraph convolution layer to obtain a dynamic hypergraph attribute representation of the node, including: Normalizing the dynamic hypergraph topology according to the hypergraph Laplacian operator to form a multi-order neighbor topology embedding; Generate a topological attribute matrix of multi-order neighbor nodes by using the multi-order neighbor topology embedding; Adaptively fuse the topological attribute matrices of all scales to form a multi-scale node attribute matrix from a multi-scale topological perspective; For any hypergraph convolutional layer in the hypergraph, the topological attribute matrix of the current convolutional layer and the multi-scale node attribute representation of the previous convolutional layer are concatenated until all hypergraph convolutional layers in the hypergraph are concatenated to obtain the dynamic hypergraph attribute representation of all nodes.

9. The drug-microorganism association prediction method based on multi-view variational graph autoencoder and multi-scale hypergraph convolution according to claim 1, characterized in that: According to the heterogeneous graph attribute representation and the dynamic hypergraph attribute representation, graph comparison learning is performed on any two graphs among the first heterogeneous graph, the second heterogeneous graph, and the hypergraph of the multi-scale topological perspective, and a drug-microorganism association prediction model is trained, including: A first loss function of the first heterogeneous graph and the dynamic hypergraph, and a second loss function of the second heterogeneous graph and the dynamic hypergraph are calculated according to the heterogeneous graph attribute representation of the node and the dynamic hypergraph attribute representation of the node; A third loss function is calculated based on the heterogeneous graph attribute representation of the nodes in the first heterogeneous graph and the attribute representation of the nodes in the second heterogeneous graph; A joint contrast loss function is calculated according to the first loss function, the second loss function and the third loss function, and a drug-microorganism association prediction model is obtained according to the joint contrast loss function.

10. A drug-microbe association prediction system based on multi-view variational graph autoencoder and multi-scale hypergraph convolution, characterized in that: include: A data acquisition module, used to acquire a drug-microorganism association database; A multi-view variational graph autoencoder module is used to construct a first heterogeneous graph based on drug function similarity and a second heterogeneous graph based on drug attribute similarity containing drug and microorganism nodes according to the drug-microorganism association database, establish a local neighbor enhanced attribute embedding, and obtain a heterogeneous graph attribute representation of the node based on the local neighbor enhanced attribute embedding and the multi-view variational graph autoencoder; A dynamic hypergraph module, used for generating a drug-microorganism hypergraph according to the drug-microorganism association database, adopting a cross-attention strategy, and constructing a dynamic hypergraph topology based on a joint hyperedge attribute matrix and a joint node attribute matrix of the hypergraph; A multi-scale convolution module, used to introduce a hypergraph Laplacian operator into the dynamic hypergraph topology to form a multi-order neighbor topology embedding, adaptively fuse the neighbor topology embedding to obtain multi-scale node attributes, and supplement the multi-scale node attributes in the hypergraph convolution layer to obtain a dynamic hypergraph attribute representation of the node; A comparative learning module, used for performing graph comparative learning on any two graphs among the first heterogeneous graph, the second heterogeneous graph and the hypergraph of multi-scale topological perspective according to the heterogeneous graph attribute representation and the dynamic hypergraph attribute representation, and training to obtain a drug-microorganism association prediction model; The association prediction module is used to obtain the characteristic vectors of drugs and microorganisms in real time, and analyze the association between the drugs and microorganisms through the drug-microorganism association prediction model.

Images