Answers are grounded in this paper and cite their sources.
Predicting individual disease trajectories and treatment responses remains challenging in drug discovery. Molecular profiling provides static snapshots that miss cellular dynamics, spatial context and adaptive responses. The convergence of single-cell and spatial omics with generative AI is driving a paradigm shift toward AI-powered virtual cells that model cellular states, interactions and treatment responses. This review introduces a nested framework of six capabilities for AI-powered virtual cells, spanning population-derived models, digital twins and agentic twins. Virtual cells generate synthetic cellular states and predict responses from population data. Digital twins link models to specific biological counterparts, enabling longitudinal updating and prospective prediction. Agentic twins add spatial context, biologically constrained interactions and actions, and environmental feedback, enabling personalized, dynamic and interactive representations of cellular ecosystems. This narrative review is based on literature acquired from PubMed, Web of Science, bioRxiv and arXiv searches up to October 2026. AI-powered virtual cells could support drug discovery as foundation models that transfer biological knowledge across tasks, including data-limited settings. Their value depends on faithful extrapolation beyond training data and prospective validation against biological observations. Coupling these models to repeatedly measured experimental systems could enable personalized prediction, iterative updating and prioritization of perturbations, targets and therapeutic strategies.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: