May 22nd.unicellularLarge Language ModelBy decoding the "molecular language" of cells, it provides a new paradigm for revealing the mysteries of life and disease mechanisms. However, the existing models are limited by data size and arithmetic bottlenecks, making it difficult to realize a qualitative leap in performance.
In response to this challenge, Professor Yang Yuedong's team from Sun Yat-sen University, together with Chongqing University, Huawei, and XinGeYuan Biotechnology, relied on the computing power of the Tianhe Xingyi supercomputing system at the National Supercomputing Center in Guangzhou, to develop a new technology that can be used in the field of biomedical research.Successfully developed the world's largest single-cell based on domestic smart computing chipbase-large model CellFM.
The model innovatively integrates data from over 100 million human single cells (more than twice the size of its peers).Built a deep learning framework with 800 million parameters (8x more parameters than similar), a major breakthrough in biological representation learning and generalization capabilities across datasets.
On May 20, the relevant results were successfully published in Nature Communications, a top international academic journal. CellFM is a successful case of training large models on domestic chips and an important research progress in the field of bio-intelligent computing in China, which provides an intelligent new engine for precision medicine and drug development.
1AI Attach the paper and code address:
https://www.nature.com/articles/s41467-025-59926-5
https://github.com/biomed-AI/CellFM