scGSDR: Harnessing gene semantics for single-cell pharmacological profiling.

Area of Science:

• Computational Biology
• Pharmacogenomics
• Precision Medicine

Background:

• Single-cell sequencing reveals cellular heterogeneity's role in drug resistance.
• Computational models can predict cellular responses to drugs using existing data.
• Understanding gene semantics and signaling pathways is crucial for biological insights.

Purpose of the Study:

• To develop a computational model (scGSDR) for predicting cellular responses to drugs.
• To enhance predictive performance by incorporating gene semantics and pathway information.
• To identify key pathways and genes contributing to drug resistance phenotypes.

Main Methods:

• Integrated two computational pipelines focusing on cellular states and gene signaling pathways.
• Incorporated gene semantics into the model for improved predictive accuracy.
• Developed an interpretability module to identify resistance-related pathways and genes.

Main Results:

• scGSDR demonstrated superior predictive accuracy compared to existing methods using bulk or single-cell RNA sequencing data.
• The model successfully predicted responses to single drugs and drug combinations.
• Biological interpretability of attention scores identified relevant genes (e.g., BCL2, CCND1, PIK3CA) and pathways involved in drug resistance.

Conclusions:

• scGSDR effectively models cellular responses to diverse drugs, including combinations, by incorporating gene semantics.
• The model aids in identifying key drug resistance pathways, advancing precision medicine.
• scGSDR facilitates targeted therapy development by pinpointing potential drug-related genes.