Abstract
Drug combinations can improve cancer therapy by boosting efficacy, limiting dose-related toxicity, and delaying resistance. We present UniSyn, an interpretable multi-modal deep learning framework that transfers knowledge from monotherapy responses to enhance drug-synergy prediction. Through hybrid attention-based integration of drug and cell-line features, UniSyn supports multi-task learning and yields mechanistic insights. It generalizes robustly to unseen drug pairs and cell types, maintaining consistent performance across multiple synergy scoring metrics. Applied at scale to tumor cell lines, UniSyn captures context-specific synergy signals and prioritizes therapeutic combinations with translational potential.