A new paradigm is emerging at the intersection of artificial intelligence and experimental biology, where cells are no longer merely observed, but comprehensively modeled, queried, and predicted in silico. This essay traces how advances in single-cell and spatial measurement technologies, perturbation biology, and AI foundation models are giving rise to virtual cells: integrative models that represent, simulate, and anticipate cellular behavior. By closing the loop between data generation, prediction, and hypothesis testing, AI is transforming biology into a self-refining, interactive science, with implications for precision medicine, biotechnology, and the scientific method.

Part of the Dædalus Special Issue on AI & Science: What Is the Future of Discovery? alongside contributions from Demis Hassabis, Yann LeCun, Pushmeet Kohli, and others.

Cells are essential to understanding health and disease, yet traditional models fall short of modeling and simulating their function and behavior. Advances in AI and omics offer groundbreaking opportunities to create an AI virtual cell (AIVC), a multi-scale, multi-modal large-neural-network-based model that can represent and simulate the behavior of molecules, cells, and tissues across diverse states. This Perspective provides a vision on their design and how collaborative efforts to build AIVCs will transform biological research by allowing high-fidelity simulations, accelerating discoveries, and guiding experimental studies, offering new opportunities for understanding cellular functions and fostering interdisciplinary collaborations in open science.

High-throughput single-cell profiling provides an unprecedented ability to uncover the molecular states of millions of cells. These technologies are, however, destructive to cells and tissues, raising practical challenges when aiming to track dynamic biological processes. As the same cell cannot be observed at multiple time points, as it changes in time and space in response to a stimulus or perturbation, these large-scale measurements only produce unaligned data sets. In this Primer, we show how such challenges can be effectively addressed using the unifying framework of optimal transport theory and tackled using the many algorithms that have been proposed for the range of scenarios of key interest in computational biology. We further review recent advances integrating optimal transport and deep learning that allow forecasting heterogeneous cellular dynamics and behaviour, crucial in particular for pressing problems in personalized medicine.

Understanding and predicting molecular responses in single cells upon chemical, genetic or mechanical perturbations is a core question in biology. Obtaining single-cell measurements typically requires the cells to be destroyed. This makes learning heterogeneous perturbation responses challenging as we only observe unpaired distributions of perturbed or non-perturbed cells. Here we leverage the theory of optimal transport and the recent advent of input convex neural architectures to present CellOT, a framework for learning the response of individual cells to a given perturbation by mapping these unpaired distributions. CellOT outperforms current methods at predicting single-cell drug responses, as profiled by scRNA-seq and a multiplexed protein-imaging technology. Further, we illustrate that CellOT generalizes well on unseen settings by (1) predicting the scRNA-seq responses of holdout patients with lupus exposed to interferon-β and patients with glioblastoma to panobinostat; (2) inferring lipopolysaccharide responses across different species; and (3) modeling the hematopoietic developmental trajectories of different subpopulations.