MGCLSyn: multimodal graph contrastive learning for drug synergy prediction

Accurate prediction of drug synergy is critical for accelerating the discovery of effective combination therapies, particularly in precision oncology. However, many existing computational approaches rely on limited data modalities or employ simple fusion strategies, which restrict their ability to capture complex drug-cell interactions and limit generalisation performance. To address these challenges, we propose MGCLSyn, a multimodal graph contrastive learning framework for drug synergy prediction. MGCLSyn integrates multiple complementary data sources, including molecular graphs, drug similarity networks, fingerprint diffusion graphs, and cell-line gene expression profiles, to learn comprehensive drug and cellular representations. In addition, a cross-modal contrastive learning mechanism is introduced to align representations across different structural views, improving consistency and discriminative capability. Furthermore, a hierarchical interaction modelling module with adaptive fusion is designed to capture complex drug-drug and drug-cell relationships. Extensive experiments on two public benchmark datasets demonstrate that MGCLSyn consistently outperforms several state-of-the-art methods in terms of predictive performance and generalisation capability. These results suggest that integrating multimodal graph representations with contrastive alignment provides an effective approach for modelling drug synergy and may facilitate the development of precision combination therapies.