Virtual cell construction for artificial intelligence-driven drug discovery

Cells are the fundamental units through which genetic variation and pharmacological perturbations influence disease processes and therapeutic responses. However, cellular responses to intervention are strongly shaped by biological context, creating a central challenge for drug discovery: predicting how specific perturbations reshape cellular systems across diverse environments. Recent advances in single-cell and spatial multi-omics technologies, large-scale perturbation profiling and artificial intelligence have made such questions increasingly tractable. These developments have driven the emergence of virtual cells as integrative computational frameworks that represent cellular states, biological context and perturbation responses within unified models. In this review, we discuss the conceptual foundations and modelling paradigms underlying virtual cell systems, including cellular representation learning, multimodal integration, perturbation prediction and mechanistic inference. We further examine how these frameworks support key drug discovery tasks, including target prioritization, drug response prediction and combination therapy design, and we outline the major challenges and future directions for virtual cells as predictive systems for therapeutic discovery.