Answers are grounded in this paper and cite their sources.
Accurate prediction of synergistic anticancer drug combinations is important for candidate prioritization in drug discovery, but current methods remain limited in providing biologically structured explanations that connect molecular interactions with cell-context-dependent gene-level signals and downstream biological processes. To address this limitation, we propose SGP-Syn, a biologically informed deep learning approach that establishes a structured Structure–Gene–Process chain. To comprehensively characterize each drug, SGP-Syn integrates complementary ChemBERTa-2 and GINE representations through fine-grained token–node attention. The resulting drug-pair representation is then integrated with cell-line gene expression to derive context-dependent gene-level signals, which are propagated through a Gene Ontology-guided visible neural network to hierarchical biological process representations. Comprehensive evaluations across two drug-combination screening datasets show that SGP-Syn outperforms state-of-the-art methods and generalizes well to unseen drugs and drug pairs, while remaining competitive in unseen-cell-line settings. Beyond predictive performance, multilevel attribution analyses link prediction-relevant molecular substructures to gene-level contributions and hierarchical biological processes, revealing shared and cell-line-specific patterns and providing biologically grounded, testable hypotheses for anticancer drug-combination discovery. SGP-Syn introduces a biologically informed Structure–Gene–Process architecture for anticancer drug synergy prediction. It links prediction-relevant molecular features to gene-level contributions and higher-level biological processes, providing structured cross-level interpretations. This design combines strong predictive performance with biologically structured interpretations for drug-combination discovery and hypothesis generation.
No takes yet. Share an insight, caveat, or question.
He et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: