Tsinghua University published ProtCell, a foundation model for single‑cell proteomics trained on about 380 mln cells across 98 public datasets. The model maps cell states and infers unmeasured protein expression, intended to support drug R&D and construction of “virtual cells.” The team reproduced the full ProtCell pretraining on Moore Threads' flagship training‑inference AI card MTT S5000; primary quantitative evaluation metrics differed by only ~0.2% versus mainstream international GPU platfor

Tsinghua University published ProtCell, a foundation model for single‑cell proteomics trained on about 380 mln cells across 98 public datasets. The model maps cell states and infers unmeasured protein expression, intended to support drug R&D and construction of “virtual cells.” The team reproduced the full ProtCell pretraining on Moore Threads' flagship training‑inference AI card MTT S5000; primary quantitative evaluation metrics differed by only ~0.2% versus mainstream international GPU platforms, validating Moore Threads' Universal GPU and MUSA full‑stack software ecosystem for life‑science foundation models.