scGSDR: Harnessing Gene Semantics for Single-Cell Pharmacological Profiling

• DOI: 10.48550/arxiv.2502.01689
• Publication Date: 2025-02-02
• Authors: Yu‐An Huang; Xiyue Cao; Zhu‐Hong You; Yue-Chao Li; Xuequn Shang; Zhi-An Huang
• Keywords: Profiling (computer programming)

The rise of single-cell sequencing technologies has revolutionized the exploration drug resistance, revealing crucial role cellular heterogeneity in advancing precision medicine. By building computational models from existing response data, we can rapidly annotate responses to drugs subsequent trials. To this end, developed scGSDR, a model that integrates two pipelines grounded knowledge states and gene signaling pathways, both essential for understanding biological semantics. scGSDR enhances predictive performance by incorporating semantics employs an interpretability module identify key pathways contributing resistance phenotypes. Our extensive validation, which included 16 experiments covering 11 drugs, demonstrates scGSDR's superior accuracy, when trained with either bulk-seq or scRNA-seq achieving high AUROC, AUPR, F1 Scores. model's application extended single-drug predictions scenarios involving combinations. Leveraging known target genes, found cell-pathway attention scores are biologically interpretable, helped us other potential drug-related genes. Literature review top-ranking genes our such as BCL2, CCND1, AKT family, PIK3CA PLX4720; ICAM1, VCAM1, NFKB1, NFKBIA, RAC1 Paclitaxel confirmed their relevance. In conclusion, semantics, modeling diverse proving invaluable single combination therapies effectively identifying resistance-related thus medicine targeted therapy development.