Google DeepMind and Yale University have unveiled a new AI system that could transform cancer research by uncovering biological insights directly validated in living cells. Announced on October 15, the foundation model, C2S-Scale 27B, is among the largest and most advanced AI systems ever built to study cellular behavior.

Based on Google’s Gemma family of models, C2S-Scale 27B generated a groundbreaking hypothesis about how cancer cells interact with the immune system, potentially reshaping future therapy design. The AI can interpret the “language” of individual cells, identifying ways to make hard-to-treat, immune-evasive “cold” tumors detectable to the body’s immune system. By revealing a mechanism that can “heat up” these tumors, the system opens the door for new combination treatments in oncology.

Figure 1. Revolutionizing Cancer Therapy.

“With more preclinical and clinical tests, this discovery may reveal a promising new pathway for developing therapies to fight cancer,” said Google CEO Sundar Pichai. Figure 1 shows Revolutionizing Cancer Therapy.

Teaching AI to read the language of cells

C2S-Scale 27B was designed to reason through complex biological conditions beyond the reach of smaller models. Its goal was to identify drugs capable of enhancing immune signaling—specifically amplifying antigen presentation, which allows immune cells to recognize cancer—under very specific conditions.

Using a dual-context virtual screen, the AI analyzed over 4,000 drugs across patient tumor samples and isolated cell data. This large-scale simulation pinpointed compounds that selectively boost immune activation in the right biological settings rather than indiscriminately.

While some of the AI’s predictions involved known drugs, 10–30% were entirely new candidates with no prior connection to cancer immunotherapy [1]. Among the most notable findings was silmitasertib (CX-4945), a kinase CK2 inhibitor. The AI predicted that silmitasertib could dramatically increase antigen presentation—but only in an “immune-context-positive” environment where low levels of interferon were present. On its own, neither silmitasertib nor interferon had much effect, but combined, they could trigger a robust immune response.

Turning “cold” tumors “hot”

Yale researchers tested the AI’s prediction in human neuroendocrine cell models not included in the training data. Experimental results confirmed the hypothesis: silmitasertib alone had no effect, and low-dose interferon produced only a modest response. However, the combination led to a 50% increase in antigen presentation, effectively activating immune recognition in previously unresponsive tumors.

reference:

1. https://interestingengineering.com/health/google-deepmind-new-ai-cancer-treatment

Cite this article:

Keerthana S (2025), DeepMind’s New AI Could Accelerate Discoveries in Cancer Treatment, AnaTechMaz, pp.842