HGAlinker: Drug-Disease Association Prediction Based on Attention Mechanism of Heterogeneous Graph

摘要

Drug repositioning raises great research interests because it saves a lot of time and economic cost in drug development. Predicting drug-disease associations based on integrated multi-dimensional biological data and networks will have better biological interpretation. At present, there are analysis challenges in scalability and information fusion across biological heterogeneous networks. In this paper, we construct a drug-disease-protein three-layer heterogeneous network, and propose a model HGAlinker to predict the links of drug and disease nodes based on the attention mechanism of heterogeneous network graph. HGAlinker has excellent performance compared with other related research methods. We analyze the model parameters and prove the robustness of the model. Through the case study, we demonstrate the biological effectiveness of the model. HGAlinker's method can also be extended to other prediction of heterogeneous network.