Overview

• Google Research, in collaboration with Yale, unveiled C2S-Scale 27B, a Gemma-based foundation model for single-cell analysis that is now publicly available.
• The model screened more than 4,000 compounds across immune-context-positive patient samples versus immune-context-neutral cell-line data to identify context-specific effects.
• It predicted a context split for the CK2 inhibitor silmitasertib, forecasting increased antigen presentation only in the immune-active setting.
• Human neuroendocrine cell experiments confirmed that silmitasertib combined with low-dose interferon raised MHC‑I antigen presentation by roughly 50%, while each agent alone had limited or no effect.
• Google reports that scaling enabled emergent conditional reasoning beyond smaller models, Yale teams are probing the mechanism and testing further predictions, and the work remains preclinical with resources shared on bioRxiv, Hugging Face, and GitHub.