AI drug clears Phase IIa trial
- Insilico Medicine’s rentosertib became the first drug with both target discovery and molecule design led by generative AI to post positive Phase IIa data. - In a 71-patient idiopathic pulmonary fibrosis trial, the 60 mg dose improved forced vital capacity by 98.4 mL, versus a 20.3 mL decline. - That matters because AI drug discovery had reached humans before, but not with peer-reviewed mid-stage efficacy data and a big-pharma validation deal.
Drug discovery is the domain here — and the stakes are simple. If AI can do more than generate molecules on slides and actually help make medicines that work in people, pharma economics changes. That gap has been the big one for years. Lots of companies claimed AI could speed up early discovery, but the field had very little mid-stage human data to prove the approach held up once biology got messy. Insilico Medicine’s rentosertib is the first clean example people can point to. In June 2025, the company and its academic collaborators published positive Phase IIa results in *Nature Medicine* for idiopathic pulmonary fibrosis, or IPF — a brutal lung-scarring disease with limited treatment options. ### What is the drug, exactly? Rentosertib — also called ISM001-055 — is a small-molecule inhibitor of TNIK, a protein target Insilico says it identified with its generative AI platform before using the same system to design the drug itself. That “both parts were AI-led” detail is why this case stands out. There have been AI-assisted programs in clinics before, but this one is being framed as the first published proof-of-concept where both target discovery and compound design came from the same AI-driven workflow. (nature.com) ### Why does IPF matter so much? IPF slowly scars the lungs until breathing gets harder and harder. Doctors often track it with forced vital capacity, or FVC — basically how much air a person can forcibly exhale after a deep breath. If FVC is falling, the lungs are usually losing function. Existing drugs can slow decline, but they do not reverse the disease, and side effects can be rough. So even modest signs of stabilization matter here. (insilico.com) ### What happened in the trial? The Phase IIa study enrolled 71 patients with IPF. The topline result that got everyone’s attention was the 60 mg once-daily arm: average FVC improved by 98.4 mL over 12 weeks, while the placebo group declined by 20.3 mL. Insilico also said the drug was safe and well tolerated, with biomarker data lining up with the proposed TNIK mechanism. That is not a final approval-grade result — but it is a real human efficacy signal, not just a preclinical promise. (nature.com) ### Why is Phase IIa the important hurdle? Because this is where a lot of shiny drug ideas start colliding with reality. Phase I mainly asks whether a drug looks safe enough to keep going. Phase IIa starts asking whether it seems to help patients in the intended disease. Passing that stage does not mean the drug works well enough for approval. But it does mean the program has crossed from “interesting platform story” into “there may actually be medicine here.” (insilico.com) ### So is this really the first AI drug success? Basically, it is the first widely cited, peer-reviewed mid-stage clinical validation for a drug whose target and molecule were both discovered and designed with generative AI. The catch is that “first” claims in biotech can get slippery, because many programs use AI in some step. But rentosertib is the clearest milestone so far for the stronger claim — AI was central to both picking what to hit and building what hits it. (nature.com) ### Why are investors paying attention now? Because the field finally has something more concrete than speed claims. Insilico says AI shortened parts of the discovery process, and the company has kept expanding its pipeline. Then in March 2026, Eli Lilly struck a deal worth up to $2.75 billion for AI-developed drugs from Insilico, with $115 million upfront. That does not prove every AI-discovered drug will work. But it shows big pharma is willing to pay for the platform after seeing real clinical traction. (insilico.com) ### What’s the catch from here? A 12-week, 71-patient Phase IIa study is promising, not definitive. IPF trials get harder as they get bigger and longer. The next questions are whether the FVC signal holds up in larger studies, whether benefits persist, and how rentosertib compares with or combines with existing treatments. Drug development is full of programs that looked good in early mid-stage trials and then faded later. (cnbc.com) ### Bottom line? The news is not that AI has “solved” drug discovery. The news is narrower — and more important. AI now has a credible, peer-reviewed Phase IIa win attached to a real drug in real patients. That moves the conversation from demos to evidence. (nature.com)
