Opening new paths in aging research

Aging research has a data problem, not a data shortage. Decades of papers exist on why cells decline, but plenty of those findings are shaky, contradictory, or simply don't replicate when someone else tries. That's the mess Calico Life Sciences waded into with Google DeepMind's Co-Scientist tool, hoping an AI system could do what overworked humans often can't: read everything and tell the difference between a real signal and noise.

Matt Onsum, who leads AI/ML at Calico, and principal scientist Katherine Labbé didn't ask Co-Scientist to just summarize papers. They used it to generate testable hypotheses, then iterated with it as lab results came in, treating it more like a research partner than a search engine. Onsum has said the tool surprised Calico's own experts with its judgment calls, its ability to spot which threads in a tangled literature were actually worth pulling.

One concrete case involves the integrated stress response, a cellular defense mechanism called ISR that protects cells under pressure but turns harmful when it stays switched on too long, a pattern tied to aging and several diseases. Calico's team wanted to understand how metabolism, which shifts as people age, regulates the ISR. Co-Scientist proposed a hypothesis about that link that researchers found plausible enough to test, then helped refine the experimental design as the work progressed.

The payoff wasn't just a proof of concept. The experiments generated new findings the team considers significant enough to publish, with real implications for how ISR dysregulation contributes to disease. That's the part worth sitting with: an AI system didn't just organize existing knowledge, it pointed toward something new that biologists then confirmed at the bench.