(Photo: Sundar Pichai, CEO of Google. Credit: Wikimedia Commons)
In a major leap for cancer research, Google DeepMind and Yale University have unveiled an artificial intelligence system capable of uncovering new biological insights directly validated in living cells.
Announced on October 15, the new foundation model, C2S-Scale 27B, represents one of the largest and most sophisticated AI systems ever developed to study cellular behavior.
Built on Google’s Gemma family of models, it has generated a groundbreaking hypothesis about how cancer cells interact with the immune system—one that could reshape how future therapies are designed.
The discovery stems from the AI’s ability to understand the “language” of individual cells, identifying how to make certain hard-to-treat or “cold” tumors visible to the body’s immune system. These types of tumors typically evade immune detection, posing one of the toughest challenges for immunotherapy.
By uncovering a mechanism that helps “heat up” these tumors, DeepMind’s system could pave the way for new types of combination treatments in oncology.
“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 in a post on X.
An exciting milestone for AI in science: Our C2S-Scale 27B foundation model, built with @Yale and based on Gemma, generated a novel hypothesis about cancer cellular behavior, which scientists experimentally validated in living cells.
With more preclinical and clinical tests,…
— Sundar Pichai (@sundarpichai) October 15, 2025
Teaching AI to read the language of cells
The C2S-Scale 27B model was designed to reason through extremely complex biological conditions that smaller models could not process. Its task was to identify drugs that could boost immune signaling. Specifically, the model aimed to amplify antigen presentation, which helps immune cells recognize cancer, but only under very particular biological conditions.
To do this, the AI used what researchers call a dual-context virtual screen, analyzing more than 4,000 drugs across patient tumor samples and isolated cell data. This large-scale simulation allowed it to identify compounds that would selectively enhance immune activation in relevant biological settings, rather than across the board.
The results were surprising. While some of the AI’s hits were already known drugs, around 10 to 30 percent were entirely new candidates—substances with no previous connection to cancer immunotherapy or immune modulation.
Among the most striking findings was the kinase CK2 inhibitor silmitasertib (CX-4945). The model predicted that silmitasertib would sharply increase antigen presentation only when used in an “immune-context-positive” environment, where low levels of interferon were already present.
Alone, the drug or interferon had minimal effect, but together they could potentially trigger a significant immune response against tumors.
Turning “cold” tumors “hot”
Yale scientists put the AI’s prediction to the test in human neuroendocrine cell models, none of which were part of the model’s training data. The experimental validation confirmed the AI’s hypothesis. Treating cells with silmitasertib alone produced no change.
Low-dose interferon alone had only a modest effect. But combining the two resulted in a 50 percent increase in antigen presentation, effectively activating immune recognition where it was previously absent.
This discovery suggests that C2S-Scale 27B didn’t just process biological data. It reasoned through context, uncovering how cellular conditions determine treatment success. The finding offers a possible roadmap for tackling tumors that resist existing immunotherapies.
Researchers at Yale are now studying the exact mechanism the AI uncovered and are testing other predictions generated by the system. The collaboration shows how large-scale AI can serve as a virtual laboratory, running thousands of simulated experiments to reveal unknown relationships between drugs, cells, and immune signals.
The success of C2S-Scale 27B underscores a shift in how scientists approach biology. Instead of traditional trial-and-error methods, AI models of this scale can generate and test hypotheses at an unprecedented pace.
Ren Ecosystem