Google DeepMind’s C2S-Scale 27B AI Proposes New Cancer Therapy Hypothesis

In mid-October 2025, Google DeepMind, in collaboration with Yale University, announced a significant breakthrough at the intersection of AI and cancer biology. Their new **C2S-Scale 27B** model, built on the Gemma architecture, can analyze single-cell datasets and generate hypotheses about cancer mechanisms that were previously beyond direct experiment. In one striking example, the model predicted that the drug **silmitasertib** (a CK2 inhibitor), when paired with low-dose interferon, could “heat up” “cold” tumors by boosting antigen presentation and making tumor cells more visible to the immune system. Lab experiments on human cell lines later validated this, showing roughly 50 % increase in antigen expression under the predicted conditions. This underlines a paradigm shift: scaled biological AI models do not just interpret data, they can **invent new biology**. The team has released code, the model is available on GitHub and Hugging Face, and the findings appear in preprints for peer review.