Article · AI translation
Thanks to IT Home readers Autumn_Dream and xxy171070 for the tips!
IT Home reported on November 14 that, according to Moore Threads’ official WeChat account, Jiang Rui’s research group in Tsinghua University’s Department of Automation recently released ProtCell, a foundation model for single-cell proteomics, and completed the paper “ProtCell: a foundation model for single-cell proteomics.” The model brings together approximately 380 million cells and 98 public datasets for pretraining and evaluation. It can accurately map cell states and infer the expression of unmeasured proteins, providing essential foundational support for drug development and the construction of cutting-edge “virtual cells.”
The paper also disclosed that the team successfully reproduced the full pretraining process for ProtCell on Moore Threads’ flagship MTT S5000 AI accelerator card for both training and inference. Its main quantitative evaluation metrics differed from those on mainstream international GPU platforms by only about 0.2%, demonstrating the ability of Moore Threads’ Universal GPU and full-stack MUSA software ecosystem to support cutting-edge foundation models in the life sciences.
Publicly available information indicates that “virtual cells” aim to use AI to build computational models of how cells function, describe their current state, and accurately predict dynamic changes after interventions such as drugs and gene editing. They have become a central strategic focus of global competition in AI for Science (AI4S).
Internationally, leading research institutions and technology giants are accelerating their investment of billions of dollars and computing clusters with tens of thousands of accelerator cards to compete in this field. In China, the State Council’s “Opinions on Deepening the Implementation of the ‘AI+’ Initiative” places “AI+” science and technology first among its priority actions. Because core data such as genomic and proteomic data on which virtual-cell research and development depends are closely tied to the security of the country’s biological genetic resources, building an advanced computing foundation that is independently controllable is crucial.