In partnership with Yale University, researchers have unveiled Cell2Sentence-Scale 27B (C2S-Scale), a 27-billion-parameter AI model built to decode the language of individual cells. Developed on the Gemma family of open models, C2S-Scale represents a major step forward in applying artificial intelligence to biological discovery.

The model has already generated a novel, experimentally validated hypothesis about cancer cell behaviour — identifying a new mechanism to help make “cold” tumours more visible to the immune system, a long-standing challenge in immunotherapy.

In simulations screening more than 4,000 drugs, C2S-Scale predicted that silmitasertib (CX-4945), a CK2 inhibitor, could amplify immune signalling only under specific conditions where low levels of interferon were present. Laboratory experiments confirmed the finding: combining silmitasertib with low-dose interferon increased antigen presentation by around 50%, making tumour cells more detectable to immune defences.

The breakthrough demonstrates that larger biological foundation models can generate entirely new hypotheses, not just refine existing knowledge. Researchers say the approach could accelerate drug discovery and transform the development of combination cancer therapies.

Teams at Yale are now investigating the mechanism behind the finding and testing additional predictions from the model across other immune contexts.

 For more information, visit this page.