For years, the narrative that AI would transform drug discovery was mostly just a story. Now there is data.
Insilico Medicine's rentosertib (ISM001-055) has completed a Phase IIa clinical trial, with results published in a peer-reviewed journal. This is the first drug where both the target discovery and the drug molecule were entirely generated by generative AI to enter human trials and produce statistically significant efficacy data.
What Makes This Target So Special
Idiopathic pulmonary fibrosis (IPF) is an extremely difficult-to-treat lung disease — lung tissue continuously scars, lung function irreversibly declines, and median survival is 3 to 5 years. Existing drugs nintedanib and pirfenidone can only slow progression, not reverse it.
In 2019, Insilico's AI system Pharma.AI analyzed vast amounts of omics data, literature, and patents, identifying a target called TNIK (TRAF2 and NCK-interacting kinase) as potentially playing a key role in IPF pathogenesis. This target had not attracted much prior research attention — precisely because AI unearthed it from data, not from a hypothesis based on existing human knowledge.
From target validation to preclinical candidate nomination, Insilico took only 18 months (2019 to February 2021). The traditional path typically takes over a decade, and that does not even account for the subsequent clinical stages.
Phase IIa Data
The trial was a standard randomized, double-blind, placebo-controlled design: 71 IPF patients across 21 centers in China, 12 weeks, four arms (placebo, 30 mg once daily, 30 mg twice daily, 60 mg once daily).
The primary endpoint was FVC (forced vital capacity), a core measure of lung function. Results:
| Group | FVC Change |
|---|---|
| 60 mg once daily | +98.4 mL |
| Placebo group | Decline |
| Subgroup not on concomitant antifibrotic therapy | +187.8 mL |
In IPF, simply halting the rapid decline in lung function is considered effective; actual improvement is rare. Insilico CEO Alex Zhavoronkov commented: "We expected safety, but we did not expect such a clear dose-dependent efficacy. Seeing FVC improvement in IPF is uncommon."
Safety Concerns That Cannot Be Ignored
Seven patients withdrew due to elevated liver enzymes or abnormal liver function, four of whom were also taking another antifibrotic drug, nintedanib. The completion rate in the high-dose (60 mg) group was only 67%, compared to 88% in the placebo group.
These signals mean that in the next pivotal trial, the dosing regimen and combination strategy for rentosertib will need optimization. Looking only at efficacy data would miss half the story.
The Real Significance
Over the past few years, the pharmaceutical industry has seen many claims of AI-designed drugs, but most remained at the computer simulation stage, or AI was only involved in one step of compound optimization. What sets Insilico apart this time is that from target discovery to drug molecule, the entire process was completed by generative AI, and it has produced the first clinical result in humans with efficacy data.
The paper is published on PubMed as peer-reviewed formal data, not a company press release.
Of course, Phase IIa only proves it is worth continuing; there are still several stages before market approval. But it has turned a previously hypothetical proposition into one backed by data: an AI-discovered target combined with an AI-designed molecule can produce a therapeutic effect in humans.
On the commercial side, Insilico recently signed a contract with Eli Lilly that includes a $115 million upfront payment and up to $2.75 billion in milestone payments. This Phase IIa data is undoubtedly a positive signal for subsequent negotiations.
Sources: Insilico's rentosertib clears a phase 2a hurdle (Drug Discovery Trends); CocoLoop, A Phase 2 Readout Generates Excitement for the Potential of AI-Driven Drug Discovery (Insilico Medicine Blog)