In the realm of biomedicine, AI technology has already demonstrated its prowess in identifying medical images and analyzing gene expression. However, its capabilities are largely confined to answering static queries, such as "What is the current biological state?" It still falls short in meeting the dynamic prediction requirements of "What will transpire following proactive changes?" To tackle this challenge, researchers have introduced a framework for world models in biomedicine. This framework aims to propel the transition of AI from "recognizing life" to "simulating life" by learning the transfer patterns between biological states. This transition hinges on intervention data and longitudinal data, yet acquiring high-quality intervention data is a costly endeavor.

Presently, the model has found preliminary applications. For instance, Alibaba DAMO Academy's cellular world model, Lingshu-Cell, can forecast the transcriptomic state of cells post-perturbation. Another notable example is the Medical World Model proposed by a research team, which can simulate tumor alterations under various treatment regimens, thereby aiding in enhancing the precision of treatment selection. Nevertheless, these models exhibit significant disparities and encounter evaluation hurdles. At this juncture, they cannot supplant real experiments and are more pragmatically employed as experimental rehearsal tools to assist researchers in screening valuable experimental pathways.

Looking ahead, AI is anticipated to play a more integral role in the decision-making process preceding experiments.