GRAM-DTI: Adaptive Multimodal Representation Learning for Drug–Target Interaction Prediction

Drug target interaction (DTI) prediction is central to computational drug discovery. While deep learning has advanced DTI modeling, existing approaches primarily rely on SMILES–protein pairs, failing to exploit rich multimodal information available for molecules and proteins. We introduce GRAM-DTI, a pre-training framework that integrates multimodal molecular and protein inputs into unified representations. GRAM-DTI extends volume-based contrastive learning to four modalities, capturing higher-order semantic alignment beyond pairwise approaches. We propose adaptive modality dropout to dynamically regulate each modality's contribution during pre-training, and incorporate IC50 activity measurements, when available, as weak supervision to ground representations in biologically meaningful interaction strengths. Experiments on four datasets demonstrate that GRAM-DTI consistently outperforms state-of-the-art baselines, highlighting the benefits of multimodal alignment and adaptive modality utilization for robust DTI prediction.