An international research team including Zhejiang University, Stanford University, the University of Toronto, and ETH Zurich published a roadmap in *Nature* on July 2 for developing a ‘virtual yeast’ AI system that predicts changes inside cells when genes are altered or drugs are administered. The paper is a perspective article proposing the data, AI architecture, and automated experimental systems required for its creation, not a study announcing actual development. The proposed model divides cellular functions into 8 specialized units (genetics, metabolism, energy, stress response, etc.), each managed by a dedicated AI, with a large language model acting as the ‘overall team leader.’ When the AI identifies genes or environmental conditions difficult to predict, automated equipment cultures real yeast and analyzes them, feeding new data back to enhance predictive accuracy. To create initial training data, the team selected 12 strains from 969 genetically and functionally diverse yeast strains and applied over 200 conditions—varying carbon/nitrogen sources, temperature, and chemicals—to these strains.