Hierarchical Cross-Level Graph Contrastive Learning for Drug-Drug Interaction Prediction

Drug-Drug Interaction (DDI) prediction is crucial for various biomedical applications like polypharmacy. Recently, some graph learning-based methods achieved promising performance in DDI prediction. However, limited attention has been given to the integration of substructure information and drug relationships to capture complex DDI patterns using self-supervised learning techniques. To this end, we propose a novel hierarchical cross-level graph contrastive learning framework named HCC, aimed at capturing hierarchical structural information and hidden DDI patterns. Firstly, we construct a drug-motif interaction graph to extract semantic motifs and model complex connections among drugs and motifs. Then, we design motif- and molecule-level self-supervised tasks. One task learns the motif-driven connectivity of the drug-motif graph, while the other learns global similarity of molecular graphs. Finally, a cross-level contrastive learning module is introduced to align multi-view information. Extensive evaluation on real-world datasets demonstrates that our method outperforms existing competitors.