MAD-TI: Meta-Path Aggregated-Graph Attention Network for Drug Target Interaction Prediction

Computational identification of unknown drug-target interactions (DTI) is crucial in locating new drug treat-ments for proteins, viruses, and diseases. This work proposes MAD-TI a meta-path-based, GAT-oriented method to predict DTIs. Our proposed method uses a heterogeneous graph of drugs, targets, diseases, and side effects as the input graph. Then, it applies two graph attention networks to generate the embeddings of drugs and targets. Using the embeddings, it predicts the unknown DTIs. The results show that MAD-TI outperforms the state-of-the-art methods.