Google’s C2S-Scale 27B model, built in collaboration with Yale under the Gemma framework, processed over a billion single-cell molecular profiles to “learn the language” of cellular communication. It generated a novel hypothesis: that the drug silmitasertib could boost antigen presentation by tumor cells — specifically converting “cold” tumors into ones more visible to the immune system. Laboratory experiments in living human cells confirmed the effect, lending strong support to the AI’s predictive ability.
Beyond the discovery itself, the article emphasizes the broader significance: AI is moving from pattern recognition to hypothesis generation. The model is now open-source to accelerate collaboration and scrutiny. Still, the article tempers excitement with realism, noting that further preclinical and clinical validation is essential before any therapeutic applications.