What Makes a Virtual Cell a World Model? Three Gaps, Three Axes, and a Roadmap

Chang Yu, Jingbo Zhou, Zhongqi Wang, Zhen Lei, Stan Z. Li

Keywords: AIVC, Virtual Cell, World Model

TL;DR: This paper defines when a virtual cell should truly qualify as a world model.

Abstract: Virtual-cell research increasingly combines mechanistic simulators, foundation models, and perturbation predictors. Yet the field often applies world-model language to systems with very different commitments. We introduce virtual cell world models (VCWMs) as a structured perspective for judgment, not a naming claim. This perspective isolates three recurring overclaims: representation is not dynamics, prediction is not intervention, and multimodality is not multiscale world modeling. We formalize a VCWM as a tuple over observations, maintained state, context, interventions, time-varying structure, transition, decoding, uncertainty, and evaluation. Assessment is compressed into three diagnostic axes: whether a model rolls out intervenable trajectories, whether it couples scales rather than only fusing modalities, and whether biological structure constrains transitions rather than only annotating embeddings. We further stage progress through capability gates from candidate VCWMs to multiscale and interactive systems. Together, these may provide a principled basis for judging when a virtual-cell system qualifies as a world model.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 38