A knowledge-driven framework for predicting single-cell responses for unprofiled drugs

Feng et al. present MAP, an AI framework that integrates biological mechanism knowledge to predict how individual cells respond to chemical perturbations, including drugs never previously profiled. The system demonstrated improved generalization to untested compounds compared to existing methods and was applied to prioritize cancer drug candidates through virtual screening. Published in Nature Machine Intelligence, the work addresses a key bottleneck in computational drug discovery — predicting cellular responses at single-cell resolution for novel molecules without requiring prior experimental perturbation data.