AI System Co-Scientist Spots Mechanism Behind MASH Drug's Limits

An AI literature-analysis system called Co-Scientist has produced a verified hypothesis explaining why the newly approved liver drug resmetirom helps only a narrow subset of the patients eligible to receive it.

The work comes from the University of Edinburgh, where bioengineer Filippo Menolascina's team applied Co-Scientist to metabolic dysfunction-associated steatohepatitis, or MASH — a common liver disease. The system proposed that the NLRP3 inflammasome acts as the specific molecular bridge coupling inflammation and metabolism in the disease.

That connection had never been pulled together into a single, actionable explanation before. Lab experiments later verified the hypothesis. According to the source, the finding could pave the way for targeted dual-therapies that pair resmetirom with a second drug acting on the inflammatory pathway.

Why MASH resists single-target drugs

MASH presents a hard problem for drug development. The disease involves intertwined biological processes — liver inflammation and metabolism — which means single-target drugs fall short. That reality pushes researchers toward combination treatments.

But combination therapy carries its own burden: the number of potential drug pairings is overwhelming. Menolascina faced that combinatorial explosion directly and used Co-Scientist to narrow the search.

The stakes are concrete. Resmetirom is a recently approved treatment prescribed for a specific stage of MASH, yet it only helps a narrow slice of those eligible patients. Understanding why has direct implications for how physicians stratify patients and how companies design next-generation regimens.

What Co-Scientist actually did

The context for the tool is the sheer volume of modern biomedical output. Biomedical research produces a flood of information that no scientist can realistically absorb, as the source puts it. Co-Scientist is built to comb that literature for overlooked links and generate new hypotheses.

In Menolascina's hands, the system did three things, according to the source. It synthesised evidence across liver biology and pharmacology. It highlighted mechanisms worth focusing on. And it flagged candidate combination therapies that his team could then take into the lab and test.

The resmetirom question is the emblematic case. Menolascina fed the system a live, practical problem: why does an approved drug, given to patients at a specific stage of MASH, work for so few of them? Co-Scientist returned a mechanistic answer — the NLRP3 inflammasome as the coupling point between the inflammatory and metabolic arms of the disease — rather than a statistical correlation.

Why this matters beyond one lab

The result is a working template for AI-assisted discovery in a field where the bottleneck is not data but hypothesis generation. In a disease defined by interacting pathways, the useful output is not another single-target candidate but a mechanistic rationale for combining two.

The Edinburgh case also shows a specific division of labor. The AI system synthesized literature, ranked mechanisms, and proposed pairings. The human team formulated the question and ran the experiments that verified the NLRP3 hypothesis. Menolascina's role was to identify which questions mattered — starting with the resmetirom puzzle — and to convert the system's output into testable science.

For drug developers, the immediate so-what is the dual-therapy angle. If resmetirom's efficacy is gated by an inflammasome-linked mechanism, then pairing it with an agent targeting that pathway becomes a rational design strategy rather than trial-and-error pairing across a combinatorial space.

Menolascina's team now has both a validated mechanism and a shortlist of candidate combinations that Co-Scientist flagged for testing. The next step, per the source, is moving those candidates through experimental validation — work that will show whether literature-mining hypotheses can consistently survive contact with the bench.