In a joint effort, Westlake University and Alibaba DAMO Academy have successfully developed an innovative stem cell AI model, christened 'Guiyuan'. This cutting-edge model leverages a large-scale combinatorial perturbation dataset, adopts a dual-modal encoding strategy, and integrates an interpretability module. These features collectively enable it to simulate and predict nearly 4 million potential stem cell reprogramming combinations. The research team undertook experimental validation using the model's optimal plan, successfully generating high-quality hypoblast-like stem cells that outperform those previously achieved. This significant breakthrough holds promise for providing a reliable cell source for investigating early human embryonic development, and facilitating in-depth research in areas such as in vitro hematopoiesis, embryo-like structure construction, and cell therapy.