In a breakthrough that could reshape cancer treatment, Google’s DeepMind has developed an artificial intelligence model that may help the body’s immune system spot and attack hidden cancer cells. The AI, built in collaboration with Yale University, discovered a way to turn so-called “cold” tumors, those that evade immune detection, into “hot” ones that the body can fight. This isn’t just a theory. The AI’s prediction was later confirmed through lab experiments on living cells, marking what Google calls “a milestone for AI in science.” What did Google’s AI actually discover
The new system, called ‘Cell2Sentence-Scale 27B (C2S-Scale 27B)’, was designed to understand the complex “language” of cells. It analysed vast amounts of biological data to learn how cancer behaves at the single-cell level. Through this process, the AI generated a completely new scientific hypothesis: that a specific drug could help expose “cold” tumors to the immune system, making them easier to attack. Researchers then tested the idea in the lab, and it worked. ‘Cold’ vs. ‘Hot’ tumors explained In cancer science, tumors are often described as cold or hot depending on how well the immune system can detect them. DeepMind’s AI essentially found a way to ‘heat up’ cold tumors, which could make previously untreatable cancers more responsive to immunotherapy. How DeepMind’s AI works The C2S-Scale 27B model, built on Google’s open-source Gemma family, contains 27 billion parameters, making it one of the largest biological AI models ever created. Its goal was to study patient tumor data and simulate how thousands of different drugs might affect immune visibility. The AI virtually screened more than 4,000 drug candidates and pinpointed one with unique potential: silmitasertib (CX-4945), a kinase CK2 inhibitor. It predicted that silmitasertib, when combined with small doses of interferon (a molecule that boosts immune signaling), could significantly increase the immune system’s ability to detect tumors. Tested and proven in living cells Yale scientists put the AI’s prediction to the test. They treated human neuroendocrine tumor cells with silmitasertib, with and without interferon. Here’s what they found: This combination effect had never been documented before, proving that DeepMind’s model didn’t just repeat known science, it generated an entirely new, testable idea. Why this discovery matters Cancer research is often slow and expensive, relying heavily on trial and error. DeepMind’s breakthrough shows how AI can act as a “virtual laboratory”, running thousands of simulated experiments and uncovering hidden biological relationships that humans might miss. As DeepMind explained: The true promise of scaling lies in the creation of new ideas, and the discovery of the unknown. If future studies confirm these findings in animals and humans, the approach could transform how scientists develop combination therapies, especially for cancers that resist existing treatments. A new chapter for AI in medicine While the discovery is still early and needs further clinical testing, it represents a turning point for AI in healthcare.
By teaching machines to understand how cells communicate, researchers can accelerate drug discovery and open new paths for treating complex diseases like cancer. As Sundar Pichai said, this could be the start of “a promising new pathway for developing therapies to fight cancer.”
