Tsinghua Releases ProtCell, an AI Model for Single-Cell Proteomics
"Virtual cells" are becoming a key battleground in global AI for Science. The idea is to use AI to build a model of how a cell works and predict how it changes after drugs or gene editing.
A team led by Jiang Rui from Tsinghua University's Automation Department recently released ProtCell, a foundation model for single-cell proteomics. It was trained and tested on about 380 million cells from 98 public datasets, and can map cell states and predict protein levels never directly measured.
The team completed the full pretraining on Moore Threads' MTT S5000 AI card. Its main test scores differed from mainstream GPU platforms by only about 0.2%, showing that domestic GPUs and the MUSA software stack can handle life science models.
The work supports drug development and virtual cell research. Since such models rely on genome and proteome data tied to biosecurity, self-controlled computing power matters more than ever.
Comments
Loading comments…