The paper introduces VCWorld, a novel "white-box" biological simulator designed to predict how cells respond to drug perturbations. Unlike traditional "black-box" AI models that lack transparency, VCWorld integrates a biological world model with the reasoning power of Large Language Models to provide interpretable, step-by-step mechanistic explanations. The system utilizes a structured knowledge graph—combining data from sources like DrugBank and Gene Ontology—alongside a specialized retrieval mechanism to anchor its predictions in established scientific facts. To support this framework, the authors also developed GeneTAK, a new benchmark dataset focused on gene-centric responses to hundreds of small-molecule drugs. Experimental results demonstrate that VCWorld achieves state-of-the-art accuracy while maintaining high data efficiency and scientific credibility. Ultimately, the research aims to accelerate drug discovery by offering a transparent computational tool for simulating complex cellular behaviors.

References:

• Wei Z, Ma R, Wang Z, et al. VCWorld: A Biological World Model for Virtual Cell Simulation[J]. arXiv preprint arXiv:2512.00306, 2025.