In a landmark announcement, Google CEO Sundar Pichai revealed that Google DeepMind’s latest artificial intelligence model has helped identify a promising new pathway for cancer therapy. Developed in collaboration with Yale University, the system builds on Google’s Gemma family of open models.
Named Cell2Sentence-Scale 27B (C2S-Scale), this 27-billion-parameter model is designed to interpret the “language” of individual cells. It generated a hypothesis about how certain harder-to-treat tumors—so-called “cold” tumors that evade immune detection—could be rendered visible to the body’s immune system. In lab tests on living human cells, the AI-predicted drug combination succeeded in significantly enhancing antigen presentation, making the tumors more recognizable to immune cells.
This accomplishment marks a shift in the role of AI from data analyzer to scientific collaborator: rather than only processing known information, the model generated a brand-new hypothesis that was experimentally validated. Researchers say this development opens the door to faster discovery of therapeutic strategies, particularly for cancers that currently resist existing treatments.
Though the results are preliminary and will require extensive pre-clinical and clinical testing, the team sees this as a potent proof of concept. With the model and its resources made available to the broader scientific community, the hope is that other researchers can build upon the work and accelerate progress in oncology.
