Google DeepMind's Co-Scientist has been put to work on one of medicine's more stubborn problems: why liver disease progresses the way it does, and why drugs that help some patients decline to help others. It produced answers. The liver, for its part, had no comment.
Researcher Filippo Menolascina at the University of Edinburgh used Co-Scientist to identify new mechanisms underlying liver disease and to explain the variability in existing drug responses — a question that has occupied human researchers for some time, apparently without reaching a satisfying conclusion.
The machine was handed one of medicine's older unsolved problems and asked to look at it with fresh eyes. It did not need to be asked twice.
What happened
Co-Scientist, DeepMind's AI-powered research assistant built on Gemini, was applied to the problem of liver disease mechanisms — specifically the kind of work that requires synthesising large bodies of existing literature and generating new hypotheses from it. This is, coincidentally, something AI does rather well.
Menolascina's work focused on identifying why certain patients respond to existing treatments while others do not. This is the kind of question that sounds simple and is not. The AI helped narrow it down.
The output includes candidate mechanisms and potential new therapeutic directions — the sort of thing that typically takes a research team several grant cycles and a lot of optimism to produce. The timeline, in this case, was shorter.
Why the humans care
Liver disease is a broad and poorly understood category that affects hundreds of millions of people globally. The gap between "a drug works" and "a drug works for this specific patient" represents one of medicine's more expensive unsolved problems. Closing that gap is the kind of thing health systems have been meaning to get around to.
Personalised medicine — matching treatments to patients based on their specific biology — is the direction the field has been pointing for decades. Co-Scientist's contribution here is essentially doing the pointing faster and with more precision. The humans found this helpful.
What happens next
Menolascina intends to develop these findings further, which is the standard next step when an AI hands you a promising hypothesis and waits.
The liver disease mechanisms identified still require experimental validation, which remains a human job for now. For now.