BGAT: A Multi Information Fusion Drug Repurposing Framework Based on Graph Convolutional Network

Abstract

Traditional drug research and development is time-consuming, expensive and low success rate. Computational drug repurposing method can find the possible drug-disease associations quickly and systematically, which is of great significance for clinical research. In recent studies, computational drug repurposing is regarded as the prediction of drug-disease link. The biological function is more and more used to interpret biological significance. According to our research, biological function data has not been used in the research of drug repurposing, but it has practical research significance.Therefore, we implement an information fusion model BGAT based on drug/disease-target, protein-biological function and PPI. BGAT model uses the fusion of multiple bipartite graph convolution networks to effectively fuse various types of data information, and deeply extract protein features to update the hidden embedding representation of drugs nodes, disease nodes and biological functions nodes. Then the BGAT model scores the drug-disease pair through the improved multilayer perceptron BMLP to accurately predict the drug-disease associations. The superiority and practicability of our model are verified by comparing with the existing dominant algorithms BiFusion, NeoDTI, and baseline algorithms that include SVM and random forest.