Drug-drug Interaction Prediction Method Based On Graph Neural Network

With the rapid development of biopharmaceutical technology,predicting the interactions between drugs has become a key point in pharmacokinetic research.Traditional artificial drug experiment methods are costly and time-consuming,while deep learning with drug big data has gradually become an effective way to predict drug interactions due to its advantages of high-speed,efficiency and low cost.Hence,drug interaction prediction models based on drug knowledge graph and deep learning have attracted more and more attention.First,we propose a drug-drug Interaction Prediction algorithm based on Graph Convolution Networks,which constructs the drug knowledge graph and drug-drug interaction dataset by using the knowledge of pharmacology,chemistry and clinical data.Furthermore,the graph convolution networkst are introduced to extract the features of high-order neighbor nodes in the drug knowledge graph,and fuse the features of high-order nodes to optimize the semantic feature of the drug node.Finally,tthe binary classifier are utilized to train the drug-drug interaction model.Experiments on drug data sets showed that the proposed model can improve the accuracy of drug interaction prediction.Second,in order to address the limitation of excessive smoothness with the increase of the convolution layer in graph convolution networks,we propose a drug-drug interaction prediction algorithm based on graph sampling attention mechanism.A random walk graph sampling is applied to construct the knowledge subgraph,which can effectively alleviate the problem of neighbor explosion and enhance the generalization ability to extract the semantic feature of the drug node.Meanwhile,the importance of nodes features to drug node can be learned by the attention mechanism,which can further improve prediction accuracy of drug-drug interaction.Finally,multiple experiments were carried out on seven public drug datasets to compare with the sate-of-the-art methods.Experimental results show that our proposed algorithm outperforms other algorithms in the term of prediction accuracy of drug-drug interaction,which can verify the effectiveness and superiority of our proposed algorithm.