A World Model of the Virtual Cell

Eric P. Xing, Le Song

Abstract: The outlook of an AI-driven digital organism, such as a virtual cell, has recently captivated much excitement and imagination from both AI and Biology communities. With a virtual cell, one can anticipate a paradigm shift of cell biology research from trial-and-error experimental exploration with cell-culture models in a wet lab, to systematic simulation of any combinatorial interventions with a computational model in a digital lab. But what constitutes an adequate realization of virtual cell? In this paper we propose an operational definition of the virtual cell based on World Model — a modern architecture recently emerged in AI research that supports advanced capabilities such as action-conditioned simulation, counterfactual reasoning, and long-horizon planning in complex dynamic environments. A world model is an AI-driven generative software system that outputs all world-possibilities upon action-prompts for simulative reasoning. When applied to biological scenarios, a world model of the virtual cell is a generative model that simulates biological possibilities of a cell under any natural or artificial interventions of the cell, or a cell population (within a tissue type or an organ). A virtual cell world model (VCWM) contrasts predictive foundation models on specific tasks, such as gene-expression perturbation prediction, as seen in some recent definitions of the virtual cell. We present a novel architecture for such a world model that enables simulated cell as an end-to-end platform: from actionable biological prompts to anticipated outcomes at all levels — molecular, structural, interactional, and morphological, in a fully aligned, integrative, multi-modal, and multi-scale fashion. We envisage that the VCWM paradigm will not simply accelerate biological experimentation, but may transform how biological possibility is explored, shifting discovery from exhaustive experimental search to structured navigation within learned cellular worlds, bringing biology closer to an industrial age of predictive design and programmable systems.