Recently, Westlake University and Alibaba DAMO Academy jointly developed a stem cell AI model called "Guiyuan". This model constructed a large-scale combinatorial perturbation dataset for stem cell reprogramming, aiming to solve the long-standing pain points in this research area, such as reliance on individual experience, repeated trial and error, and high costs and low success rates of traditional experimental verification.

Stem cell reprogramming involves 25 lineage regulatory factors, theoretically forming nearly 4 million different potential combinations. The "Guiyuan" model adopts a dual-modal encoding strategy, which can accurately predict the impact of different combinations on cell fate and innovatively includes an explainability module.

Currently, the "Guiyuan" model has completed the simulation prediction of nearly 4 million potential combinations. The research team conducted experimental validation based on the optimal solutions provided by the model and successfully obtained high-quality embryonic-like stem cells, whose quality is better than previous results.

This breakthrough is expected to provide a reliable source of cells for subsequent applications, not only helping to understand human early embryonic development but also further promoting in-depth research in areas such as in vitro hematopoiesis, embryo-like construction, and cell therapy.