Overview

• Google Research/DeepMind and Yale unveiled Cell2Sentence-Scale 27B, a Gemma-based foundation model for single-cell analysis.
• The model conducted a dual-context virtual screen of more than 4,000 drugs, contrasting patient tumor samples with immune activity against isolated cell-line data to find context-dependent amplifiers.
• C2S-Scale predicted that the CK2 inhibitor silmitasertib would increase antigen presentation only with low-level interferon signaling.
• Yale-linked lab tests in human neuroendocrine cells confirmed the prediction, showing about a 50% rise in antigen presentation with silmitasertib plus low-dose interferon, while each agent alone had limited effect.
• The model, preprint, and code are publicly available on bioRxiv, Hugging Face, and GitHub, and Yale teams are probing mechanism and testing further AI-generated predictions as the findings await peer review.