An AI-Designed Drug Reached Phase 3 — and So Did Its Target

Insilico Medicine has dosed the first patient in GENESIS-IPF-3, a Phase III trial of rentosertib for idiopathic pulmonary fibrosis. The study is a 52-week prospective, randomized, multi-center, double-blind, placebo-controlled, parallel-group design, enrolling 320 participants across 47 centers in China. The primary endpoint is the annual rate of decline in forced vital capacity over 52 weeks.

Insilico's claim is that this is the first drug generated by generative AI — against a target also identified by AI — to reach a Phase III trial.

Both halves of that claim need separating, because only one of them is genuinely novel.

Generated molecule versus discovered target

AI-designed molecules entering clinical trials is no longer remarkable. Generative chemistry has been producing candidates for years, and several have reached human testing. Designing a molecule is, in computational terms, a search problem over a well-defined space with well-defined scoring functions. It is hard, and it is tractable.

Picking the target is a different kind of problem. Target identification asks which protein, out of roughly twenty thousand, is causally upstream of a disease in a way that a small molecule can usefully perturb. It is the step where most drug programs fail, and the failure is expensive — you find out in Phase II, after eight years and several hundred million dollars, that the biology was wrong.

Rentosertib inhibits TNIK, a protein linked to the signaling pathways driving lung scarring, chronic inflammation and cellular aging. Insilico's position is that TNIK was surfaced by its target-discovery platform rather than selected from existing literature consensus.

If that holds through Phase III, it is the first time an AI-nominated novel target has been validated in a large randomized trial. That is a substantially bigger claim than "AI drew the molecule," and it is the one worth tracking.

Why IPF, and why the endpoint is honest

Idiopathic pulmonary fibrosis is a progressive scarring of the lungs with no known cause, producing breathlessness and chronic cough, and it is fatal. Existing therapies slow decline; none reverse it.

That makes it an unusually clean setting for a novel-mechanism trial. The standard of care is weak enough that a placebo-controlled study is ethical, the disease progresses fast enough that 52 weeks is sufficient to see separation, and the primary endpoint — FVC decline — is an objective, well-validated spirometry measurement rather than a composite or a patient-reported score.

Phase IIa results showed patients on a 60 mg daily dose with improved lung function versus placebo. That is encouraging and it is a Phase IIa, which means small, and which means the effect size is the thing Phase III exists to test.

The longevity result, and why to hold it loosely

On September 7, Insilico published a study in Nature Biotechnology reporting that blood samples from 42 rentosertib patients showed signs of reversed biological age.

Insilico's own executives cautioned that the effect is small and may not be sustained over a longer period. That caveat is doing a great deal of work and deserves to be repeated rather than buried.

Biological-age measures derived from blood are correlational proxies. Forty-two patients is a small sample. "Signs of reversal" in a secondary analysis of a Phase IIa cohort is a hypothesis-generating observation, not a finding — and TNIK's connection to cellular aging pathways makes it exactly the kind of result that is mechanistically plausible and therefore especially easy to over-read.

The company published the caveat alongside the claim, which is the correct behavior. The risk is that the caveat does not travel as far as the headline.

The China-first path

Running the pivotal trial across 47 Chinese centers is a strategic choice with real tradeoffs.

The advantages are speed and cost. Chinese trial enrollment for a condition like IPF is faster than in the US or EU, per-patient costs are lower, and China's regulator has moved considerably on timelines for innovative therapies. For a company that needs a Phase III readout to validate an entire platform thesis, time-to-data is the dominant variable.

The disadvantage is regulatory portability. An FDA or EMA filing on a single-country Phase III dataset is a harder conversation, and one that usually requires bridging studies or a multiregional design. A positive result in China is a genuine approval pathway in China and a strong but incomplete package elsewhere.

Insilico has been explicit, through founder Alex Zhavoronkov, that the bet is on China plus AI delivering the industry's next breakthrough. This trial is that bet made concrete.

What to watch

The readout, roughly 52 weeks after enrollment completes. Everything above is prologue. FVC decline versus placebo is the number.

Whether the TNIK mechanism holds. A positive trial on a novel AI-nominated target changes how pharma values target-discovery platforms. A negative one on a valid molecule tells you generative chemistry works and target selection does not — which is the outcome the industry should be most prepared for.

Whether a Western regulator engages early. A pre-submission conversation with FDA about this dataset would signal Insilico is building toward a global filing rather than a domestic one.

The longevity claim's second dataset. Forty-two patients is a signal to check. If it does not replicate in the Phase III cohort, it should be retired quietly rather than repeated.