TriM-DTA: A Tri-Modal Fusion Framework for Drug-Target Binding Affinity Prediction

The process of discovering new therapeutic drugs is often time-consuming and resource-intensive, making the precise estimation of the binding strength between drug candidates and their target proteins an essential step in modern computational pharmacology. Traditional methods mostly rely on molecular data from a single modality such as sequence and structure, making it difficult to effectively obtain the spatial and biochemical characteristics of complementarity in molecular interactions. In this work, we propose TriM-DTA, a tri-modal information fusion framework to accurate predict drug-target binding affinity that integrates sequence features, topological graphs, and geometric structures of both drugs and targets. Through a dedicated encoder for each modality and a hierarchical fusion scheme, our model achieves a more holistic understanding of drug-target complexes, as evidenced by its competitive performance on benchmark datasets. Ablation studies confirm the distinct contribution of each module to overall performance.