Answers are grounded in this paper and cite their sources.
Cancer drug response prediction (DRP) remains a challenging problem due to the intrinsic heterogeneity of tumors and the dependence of structure-activity relationships on cellular context. Most existing approaches rely on single-omics data and global molecular representations, limiting their ability to capture how specific chemical substructures, known as motifs, interact with complex cellular states. We present C ell- C onditioned M otif G raph (CCMG), a graph-based framework that explicitly models interactions between molecular motifs and multi-omics cellular context for accurate and interpretable DRP. CCMG augments molecular graphs with motif-level representations and integrates them with cell-specific multi-omics embeddings through a unified graph structure, enabling context-aware learning of structure-activity relationships. This design allows the model to capture fine-grained substructure-cell interactions while enabling interpretability at the atom, motif, and gene levels. We evaluate CCMG on the Genomics of Drug Sensitivity in Cancer (GDSC) under several data-splitting schemes and conduct ablations to quantify each component’s contribution. Results demonstrate that explicit motif-cell coupling improves generalization over late-fusion baselines, particularly under challenging splits, while offering mechanistic insights that support downstream biological interpretation and clinical hypothesis generation.
No takes yet. Share an insight, caveat, or question.
Nguyen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: