Drug-Disease Association Prediction Based on Meta-Path Heterogeneous Network with Global Graph Attention

In this paper, a heterogeneous network model based on global graph attention meta-path, named MHNGA, is proposed for drug-disease association prediction. Firstly, the data of drugs and diseases are collected, and the known drug-disease association, drug similarity and disease similarity are constructed as a heterogeneous network. Secondly, multiple meta-path-based subgraphs are introduced, and the graph attention neural network is used to extract the features of the neighbor nodes of these subgraphs, and the features are enhanced by channel attention and spatial attention mechanisms. Finally, through the evaluation of ten-fold cross-validation, MHNGA achieves 93.5% of the area under the accurate recall curve and 99.4% of the accuracy. © 2024 University of Electronic Science and Technology of China. All rights reserved.