Accurately modeling drug synergy across diverse biological contexts remains a key challenge in computational pharmacology, especially as combination therapies gain prominence in the era of precision medicine. Effective prediction of synergistic interactions requires capturing not only the intrinsic properties of individual molecules, but also how these properties interact in a cell-type specific environment. To address this, we introduce SynerGAT, a deep learning framework that models both intra- and inter-molecular dependencies while seamlessly integrating biologically meaningful context. By combining graph-based molecular representations with sequence-level modeling and a novel cross-attention mechanism, SynerGAT learns interaction-aware embeddings that adapt dynamically to different cellular conditions. The integration of contextual information enables the model to generalize beyond fixed chemical structures and account for variability in biological response. [11] Empirical evaluations on benchmark synergy prediction datasets demonstrate that SynerGAT outperforms established baseline models. SynerGAT achieves a high predictive accuracy with a ROC-AUC of 0.97, outperforming or matching leading baselines such as DKPE-GraphSYN and MatchMaker. Beyond predictive performance, its architectural novelty—namely the dual-pathway attention design—contributes to model interpretability and modular extensibility, making SynerGAT a biologically informed tool for modeling drug synergy in complex settings.