Gemma Model Helps Discover New Cancer Therapy Pathway

Google DeepMind and Yale University have unveiled Cell2Sentence-Scale 27B (C2S-Scale), a 27-billion-parameter foundation model built on the Gemma family to decode the “language” of individual cells. The model achieved a breakthrough in cancer research by predicting that the kinase inhibitor silmitasertib, when combined with low-dose interferon, could enhance antigen presentation in tumors, making “cold” tumors more visible to the immune system. This AI-generated hypothesis was experimentally validated in human cell models, revealing a potential new pathway for immunotherapy. C2S-Scale demonstrates that scaling biological models not only improves accuracy but also enables novel scientific discovery. The model and code are now publicly available for researchers via bioRxiv, Hugging Face, and GitHub.

Google DeepMind, Shekoofeh Azizi and Bryan Perozzi – How a Gemma model helped discover a new potential cancer therapy pathway – 15th October 2025