REDDA: Integrating multiple biological relations to heterogeneous graph neural network for drug-disease association prediction

Yaowen Gu, Si Zheng, Qijin Yin, Rui Jiang, Jiao Li

Abstract: Highlights•A large-scale benchmark for drug-disease association is proposed with 5 entities and 10 relations.•A Heterogeneous graph neural network is proposed for drug-disease association prediction called REDDA.•REDDA uses 3 attention mechanisms and a topological subnet to learn comprehensive node representations.•REDDA outperforms 8 advanced baselines and gain impressive improvements on 2 datasets.•Attention visualization analyzes and case study demonstrate the effectiveness of the model architecture of REDDA.