An AI-Designed Drug Reaches Patients: What Rentosertib Actually Proved

AI drug discovery has promised faster medicine for years. Rentosertib reached a randomized phase 2a trial for people with idiopathic pulmonary fibrosis.

The drug is unapproved, and the trial established a fairly narrow point: an AI-assisted search produced a candidate credible enough for meaningful human testing.

Why pulmonary fibrosis needs better options

Idiopathic pulmonary fibrosis causes progressive scarring that makes the lungs stiffer and breathing harder. Existing drugs can slow decline for some patients. They do not reverse the disease, and side effects can make treatment difficult.

Researchers used AI platforms to identify TNIK, a kinase involved in inflammatory and fibrotic pathways, as a possible target. Generative chemistry tools then helped design and optimize rentosertib, previously called ISM001-055, to inhibit it (Xu et al.).

The developers reported reaching a preclinical candidate in about 18 months. Speed is one of the clearest potential advantages of AI here: it can search possible targets and molecular structures more broadly before researchers choose what to synthesize and test.

What the phase 2a trial found

The multicenter, double-blind trial randomized 71 participants to placebo or one of three rentosertib dosing schedules for 12 weeks. Its primary endpoint was safety, measured through treatment-emergent adverse events.

Adverse events were common in every group, including placebo, and occurred at broadly similar rates. Treatment-related serious events were low and comparable. Liver toxicity and diarrhea were among the reasons some participants stopped treatment.

The highest-dose group, receiving 60 milligrams once daily, had an average increase of 98.4 milliliters in forced vital capacity. The placebo group declined by an average of 20.3 milliliters. Forced vital capacity measures how much air a person can exhale and is commonly used to follow pulmonary fibrosis.

I find that difference encouraging. It was a secondary result from a small dose group followed for only 12 weeks, and the study was designed primarily to evaluate safety. A larger and longer trial could produce a smaller effect, no effect, or a clearer benefit.

What was AI’s contribution?

AI contributed early by helping researchers prioritize a target and generate a molecule with the desired properties. Researchers then tested the biology, synthesized compounds, evaluated toxicity, selected doses and conducted the human trial.

That early search could reduce time spent on dead ends. Every candidate still faces the biological uncertainty that causes most drugs to fail.

The label “AI-designed drug” gives the software too much credit. It helped researchers decide where to look and what to build. Rentosertib still faces the same evidence process as any other drug.

A larger trial still has to show benefit

Rentosertib shows that generative-AI discovery can reach patients and produce an interpretable clinical signal. This trial does not tell us whether AI-discovered candidates succeed more often or cost less across the complete development process.

Phase 3 is where a treatment generally has to demonstrate benefit in a larger population and over a meaningful period. According to the study authors, AI-discovered drugs had not yet crossed that full threshold when the trial was published.

My Thoughts

I paid attention because the evidence comes from a randomized clinical trial. A computational search produced a candidate that reached patients, which is already an accomplishment.

I care much more about the next study than the discovery label. Rentosertib now has to preserve lung function in a larger group for longer than 12 weeks. Even a fast discovery process still ends at the same difficult test: whether changing the biology helps patients.