Drug-miRNA interaction prediction method based on meta-path expert network

Abstract

The invention discloses a drug-miRNA interaction prediction method based on a meta-path expert network, belongs to the field of biological information, and relates to a drug-miRNA interaction prediction method. Prediction of the interaction between the drug and the miRNA not only contributes to revealing the drug action mechanism, but also can provide theoretical support for drug target discovery, drug relocation and individualized treatment. A traditional method mainly verifies drug-miRNA association based on biological experiments, but has the problems of high cost and long consumed time. However, the existing calculation method does not fully excavate the relationship between the drug and the miRNA. The invention provides a drug-miRNA interaction prediction method based on a meta-path expert network, and the method comprises the steps: firstly fusing various similarities, and constructing a heterogeneous graph in combination with the known drug-miRNA interaction; a heterogeneous graph is converted into a plurality of isomorphic graphs through meta-paths, semantic information of different meta-path graphs is learned by using a graph convolutional network, and node features with rich semantics are obtained. And finally, fusing node features of different meta path diagrams by using a hybrid expert model to obtain final drug node features and miRNA node features, and performing interaction prediction through MLP.

Description

Technical Field

[0001] The present invention belongs to the field of bioinformatics, and in particular relates to a drug-miRNA interaction prediction method based on a meta-path expert network. Background Art

[0002] As a class of endogenous small RNAs, approximately 20-24 nucleotides in length, miRNAs are involved in numerous life processes, including tumorigenesis, reproduction, and metastasis. They offer new avenues for developing novel drug targets. Drugs can intervene in various disease processes by influencing miRNA expression, which holds significant promise for treatment.

[0003] Traditional experimental methods can directly verify drug-miRNA interactions, but they are time-consuming and labor-intensive. In recent years, computational methods have become mainstream. Early researchers used the ligand affinity of pre-miRNA and the protrusion structure in miRNA based on the three-dimensional structure of miRNA to search for associated drugs. To date, the rise of deep learning has provided a favorable production tool for computational methods. Researchers construct heterogeneous networks by integrating drug-miRNA associations, drug features, sequence features, etc., and explore potential associations through matrix decomposition or graph neural networks. Although these methods have achieved good results in predicting drug-miRNA interactions, they still do not fully capture the relationship between drugs and miRNAs, which has a certain negative impact on the accuracy of prediction results. Summary of the Invention

[0004] The present invention proposes a drug-miRNA interaction prediction method based on a meta-path expert network to solve the problem that the relationship between drugs and miRNAs cannot be fully captured and the regulatory effect of drugs on miRNAs is difficult to accurately reveal.

[0005] The above invention objectives are mainly achieved through the following technical solutions:

[0006] S1. Similarity calculation: Integrate and calculate the chemical structure similarity of drugs Similarity of side effects and functional similarity Sequence similarity of miRNAs Disease semantic similarity and functional similarity Generate the fusion similarity S of the drug drug and miRNA fusion similarity S mirna ;

[0007] S2. Constructing heterogeneous graphs: combining known drug-miRNA interactions and drug fusion similarity drug and miRNA fusion similarity S mirna,construct a heterogeneous graph containing drug nodes and miRNA nodes;

[0008] S3. Convert heterogeneous graphs: Design meta-paths for drug nodes and miRNA nodes respectively, and convert the heterogeneous graphs into multiple drug meta-path graphs. Multiple miRNA metapathway maps

[0009] S4. Meta-pathway graph node feature learning: Using graph convolutional networks to learn node features of different drug meta-pathway graphs Node characteristics of miRNA metapathway graph

[0010] S5. Hybrid Expert Feature Fusion: Concatenating Node Features of Multiple Drug Metapathway Graphs get Node features of splicing multiple miRNA metapathway graphs get Use hybrid expert model to process and By aggregating the outputs of different experts to achieve internal feature fusion, the final drug node features are obtained. and miRNA node characteristics

[0011] S6. Interaction Prediction: Splice Drug Node Features and miRNA node characteristics The artificial neural network was input to obtain the prediction score of drug-miRNA interaction.

[0012] Effects of the Invention

[0013] This paper proposes a method for predicting drug-miRNA interactions based on a meta-path expert network. This method integrates multiple similarity information to obtain effective similarity information. Meta-paths are used to fully explore the relationships and roles of drugs and miRNAs in regulatory networks. A hybrid expert model is then used to obtain high-quality node features, providing a new approach for predicting drug-miRNA interactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Overall flow chart;

[0015] Figure 2 Method structure diagram.

[0016] Specific implementation methods DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] The present invention provides a heterogeneous graph node relationship prediction method based on graph transformation network and graph decomposition, taking the relationship prediction between drug nodes and protein nodes as an example, comprising the following steps:

[0019] S1. Similarity calculation: Integrate and calculate the chemical structure similarity of drugs Similarity of side effects and functional similarity Sequence similarity of miRNAs Disease semantic similarity and functional similarity Generate the fusion similarity S of the drug drug and miRNA fusion similarity S mirna ;

[0020] S2. Constructing heterogeneous graphs: combining known drug-miRNA interactions and drug fusion similarity drug and miRNA fusion similarity S mirna ,construct a heterogeneous graph containing drug nodes and miRNA nodes;

[0021] S3. Convert heterogeneous graphs: Design meta-paths for drug nodes and miRNA nodes respectively, and convert the heterogeneous graphs into multiple drug meta-path graphs. Multiple miRNA metapathway maps

[0022] S4. Meta-pathway graph node feature learning: Using graph convolutional networks to learn node features of different drug meta-pathway graphs Node characteristics of miRNA metapathway graph

[0023] S5. Hybrid Expert Feature Fusion: Concatenating Node Features of Multiple Drug Metapathway Graphs get Node features of splicing multiple miRNA metapathway graphs get Use hybrid expert model to process and By aggregating the outputs of different experts to achieve internal feature fusion, the final drug node features are obtained. and miRNA node characteristics

[0024] S6. Interaction Prediction: Splice Drug Node Features and miRNA node characteristics The artificial neural network was input to obtain the prediction score of drug-miRNA interaction.

[0025] The present invention first fuses multiple similarities and constructs a heterogeneous graph based on known drug-miRNA interactions. Metapaths are then used to convert the heterogeneous graph into multiple homogeneous graphs. A graph convolutional network is then used to learn the semantic information of different metapath graphs, obtaining semantically rich node features. Finally, a mixture of experts model is used to fuse the node features of different metapath graphs to obtain the final drug node features and miRNA node features. These features are then concatenated and used for interaction prediction using an artificial neural network.

[0026] The embodiments of the present invention are described in detail below:

[0027] S1. Similarity calculation: Integrate and calculate the chemical structure similarity of drugs Similarity of side effects and functional similarity Sequence similarity of miRNAs Disease semantic similarity and functional similarity Generate the fusion similarity S of the drug drug and miRNA fusion similarity S mirna :

[0028] S11. Drug chemical structure data and drug function data can be downloaded from the DrugBank database (https: / / go.drugbank.com / ), and drug side effect data can be downloaded from the SIDER database (http: / / sideeffects.embl.de / );

[0029] S12. Download miRNA sequence data from the miRBase database (https: / / www.mirbase.org / ); download miRNA-related disease data from the miR2Disease database (http: / / www.mir2disease.org / 7) and the HMDD v4.0 database (http: / / www.cuilab.cn / hmdd); miRNA functional similarity It can be downloaded directly from the MISIMv2.0 database (http: / / www.lirmed.com / misim / Home);

[0030] S13. Calculate the chemical structure similarity of drugs using SIMCOMP (https: / / www.genome.jp / tools / simcomp / ) Select the top ten drug side effects and calculate the similarity of drug side effects using Jaccard similarity Based on the drug-related target data in the DrugBank database, a target gene set was constructed and the functional similarity of drugs was calculated using Jaccard similarity.

[0031] S14. Calculate sequence similarity using the Needleman-Wunsch algorithm based on miRNA sequence data According to miRNA-related disease data, Jaccard similarity is used to calculate disease semantic similarity The Jaccard similarity calculation formula is as follows:

[0032]

[0033] Among them, B and C are two sets of diseases related to miRNAs, |B∩C| is the intersection of the two, and |B∪C| is the union of the two. Similarly, the similarity of drug side effects can be calculated. Functional similarity to drugs

[0034] S15. Integrate and calculate the chemical structure similarity of drugs Similarity of side effects and functional similarity Sequence similarity of miRNAs Disease semantic similarity and functional similarity The fusion similarity S of drugs is calculated according to the following formula drug and miRNA fusion similarity S mirna :

[0035]

[0036] Among them, α i and β i is a weight parameter used to adjust the importance of various similarities; i is 1-3, and j is 1-3.

[0037] S2. Constructing heterogeneous graphs: combining known drug-miRNA interactions and drug fusion similarity drug and miRNA fusion similarity S mirna , construct a heterogeneous graph containing drug nodes and miRNA nodes:

[0038] S21, download known drug-miRNA interaction data from the SM2miR database (http: / / www.jianglab.cn / SM2miR / );

[0039] S22. Based on the fusion similarity of drugs drug Connect similar drug nodes according to the fusion similarity S of miRNA mirna Similar miRNA nodes are connected, and edges are added between drug nodes and miRNA nodes with known interactions to obtain a heterogeneous graph containing drug nodes and miRNA nodes.

[0040] S3. Convert heterogeneous graphs: Design metapaths for drugs and miRNAs respectively, and convert heterogeneous graphs into multiple drug metapath graphs Multiple miRNA metapathway maps

[0041] S31. Design metapaths for drug nodes and miRNA nodes respectively. The design rule is that the starting and ending nodes of the metapaths are of the same type. Different metapaths contain different high-level semantics. The specific metapath designs are as follows:

[0042] Drug node metapath: drug-drug, drug-miRNA-drug, drug-miRNA-miRNA-drug; miRNA node metapath: miRNA-miRNA, miRNA-drug-miRNA, miRNA-drug-drug-miRNA;

[0043] S32. Connect related drug nodes in the heterogeneous graph according to the meta-path, and convert the heterogeneous graph into multiple drug meta-path graphs. Similarly, the heterogeneous graph can be converted into multiple miRNA meta-path graphs.

[0044] S4. Meta-pathway graph node feature learning: Using graph convolutional networks to learn node features of different drug meta-pathway graphs Node characteristics of miRNA metapathway graph

[0045] S41. Learning node features of drug meta-pathway graph using graph convolutional networks and node characteristics of miRNA metapathway graph Taking the drug meta-path graph as an example, the node feature calculation formula of the l+1 layer is as follows:

[0046]

[0047] Among them, D m is the degree matrix, is the adjacency matrix with self-loops added, is the node feature of the lth layer, is the weight matrix of the lth layer, and similarly, the node features of the miRNA meta-pathway graph can be obtained

[0048] S5. Hybrid Expert Feature Fusion: Concatenating Node Features of Multiple Drug Metapathway Graphs get Node features of splicing multiple miRNA metapathway graphs get Use hybrid expert model to process and By aggregating the outputs of different experts to achieve internal feature fusion, the final drug node features are obtained. and miRNA node characteristics

[0049] S51. Node features of splicing different drug meta-pathway graphs get Node features of splicing different miRNA metapathway graphs get

[0050] S52, will and Input the mixed expert model separately, summarize the outputs of different experts according to the weights, and obtain the final drug node features and miRNA node characteristics The summary formula is as follows:

[0051]

[0052] in, The weight assigned to each expert when fusing node features of different drug meta-pathway graphs, is the weight assigned to each expert when fusing node features of different miRNA meta-pathway graphs; For the output of each expert, Output of experts.

[0053] S6. Interaction Prediction: Splice Drug Node Features and miRNA node characteristics The artificial neural network was input to obtain the prediction score of drug-miRNA interaction.

[0054] The present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of the present invention.

Claims

1. A drug-miRNA interaction prediction method based on meta-path expert network, characterized in that: The following steps are involved: S1. Similarity calculation: Integrate and calculate the chemical structure similarity of drugs Similarity of side effects and functional similarity Sequence similarity of miRNAs Disease semantic similarity and functional similarity Generate the fusion similarity S of the drug drug and miRNA fusion similarity S mirna ; S2. Constructing heterogeneous graphs: combining known drug-miRNA interactions and drug fusion similarity drug and miRNA fusion similarity S mirna ,construct a heterogeneous graph containing drug nodes and miRNA nodes; S3. Convert heterogeneous graphs: Design meta-paths for drug nodes and miRNA nodes respectively, and convert the heterogeneous graphs into multiple drug meta-path graphs. Multiple miRNA metapathway maps S4. Meta-pathway graph node feature learning: Using graph convolutional networks to learn node features of different drug meta-pathway graphs Node characteristics of miRNA metapathway graph S5. Hybrid Expert Feature Fusion: Concatenating Node Features of Multiple Drug Metapathway Graphs get Node features of splicing multiple miRNA metapathway graphs get Use hybrid expert model to process and By aggregating the outputs of different experts to achieve internal feature fusion, the final drug node features are obtained. and miRNA node characteristics S6. Interaction Prediction: Splice Drug Node Features and miRNA node characteristics The artificial neural network was input to obtain the prediction score of drug-miRNA interaction.

2. The drug-miRNA interaction prediction method based on meta-path expert network according to claim 1, characterized in that: The fusion similarity between the drug and miRNA described in step S1 is calculated as follows: S11. Integrate and calculate the chemical structure similarity of drugs Similarity of side effects and functional similarity Sequence similarity of miRNAs Disease semantic similarity and functional similarity The fusion similarity S of the drug is calculated according to the following formula drug and miRNA fusion similarity S mirna : Among them, α i and β i is a weight parameter used to adjust the importance of various similarities; i takes a value of 1-3, and j takes a value of 1-3.

3. The drug-miRNA interaction prediction method based on meta-path expert network according to claim 1, characterized in that: The steps of designing metapaths in step S3 to convert the heterogeneous graph into a drug metapath graph and a miRNA metapath graph are as follows: S31. Design metapaths for drug nodes and miRNA nodes respectively. The design rule is that the starting node and the ending node of the metapath are of the same type. S32. Connect related drug nodes in the heterogeneous graph according to the meta-path, and convert the heterogeneous graph into multiple drug meta-path graphs. Similarly, the heterogeneous graph can be converted into multiple miRNA meta-path graphs.

4. The drug-miRNA interaction prediction method based on meta-path expert network according to claim 1, characterized in that: Node features of the drug meta-pathway graph in step S4 Node characteristics of miRNA metapathway graph The steps to obtain it are as follows: S41. Learning node features of drug meta-pathway graph using graph convolutional networks and node characteristics of miRNA metapathway graph Taking the drug meta-path graph as an example, the node feature calculation formula of the l+1 layer is as follows: Among them, D m is the degree matrix, is the adjacency matrix with self-loops added, is the node feature of the lth layer, is the weight matrix of the lth layer, and similarly, the node features of the miRNA meta-pathway graph can be obtained 5. The drug-miRNA interaction prediction method based on meta-path expert network according to claim 1, characterized in that: When the node features of different drug meta-path graphs are fused in step S5, the final drug node features are obtained. and miRNA node characteristics Here are the steps: S51. Node features of splicing different drug meta-pathway graphs get Node features of splicing different miRNA metapathway graphs get S52, will and Input the mixed expert model separately, summarize the outputs of different experts according to the weights, and obtain the final drug node features and miRNA node characteristics The summary formula is as follows: in, The weight assigned to each expert when fusing node features of different drug meta-pathway graphs, is the weight assigned to each expert when fusing node features of different miRNA meta-pathway graphs; For the output of each expert, Output of experts.

Images