This post is also available in:
עברית (Hebrew)
A new artificial intelligence model developed by Google DeepMind and Yale University has shown promising results in identifying drug combinations that could improve the immune system’s ability to interact with certain types of cancer. The findings were recently validated in human cell experiments, highlighting how large-scale AI can help uncover mechanisms that might otherwise remain hidden.
The system, called C2S-Scale 27B, is one of the most advanced models built for studying cellular behavior. Designed to interpret how cells respond under specific biological conditions, it was tasked with identifying drugs that could enhance immune signalling — in particular, boosting antigen presentation, a key step that helps the immune system recognize cancer cells.
To accomplish this, the model simulated over 4,000 drug interactions across tumor samples and isolated cell data, using a method known as dual-context virtual screening. This approach allowed the AI to identify drug effects not in isolation, but in biologically relevant environments.
According to Interesting Engineering, one of the most significant findings from this process was the identification of the kinase CK2 inhibitor silmitasertib (CX-4945) as a potential enhancer of immune response — but only when used alongside low doses of interferon. On its own, neither compound had a strong effect. But when combined, the two significantly increased antigen presentation in tumor cells, effectively activating immune recognition.
The results were experimentally confirmed in neuroendocrine cell models that had not been part of the training data. The combination treatment led to a 50% increase in antigen presentation, suggesting a possible new approach for treating “cold” tumors — those that typically evade immune detection and respond poorly to existing immunotherapies.
Beyond this specific discovery, the model also flagged several other drug candidates with no previous known role in cancer immunotherapy, suggesting broader potential for future research.
This development reflects a growing shift in biomedical research, where AI is increasingly used not just to process data but to generate testable hypotheses, enabling faster exploration of complex biological systems.
The discovery was published on Google’s blog.
