AI-Derived Therapy for IPF Shows Potential in Phase 2a Trial

A novel small molecule therapy for idiopathic pulmonary fibrosis (IPF) that was first identified using artificial intelligence (AI) was found to be safe and well-tolerated in a phase 2a trial.¹ The results, published in Nature Medicine, could lead to the development of one of the first new therapies attributable to generative AI, although longer, larger-scale trials are needed to confirm the initial findings.

With all that has been written about the potential of generative AI, one of the most impactful potential uses of the technology is to discover novel therapeutic targets and design therapies based on those insights. Yet, as corresponding author Alex Zhavoronkov, PhD, and colleagues explained, the potential remains largely theoretical to date.

Zhavoronkov is the chief executive of Insilico Medicine, a global company working to use AI for drug discovery. He and colleagues noted that phase 2 trials based on AI-discovered drugs have generally had similar levels of failure compared with therapies discovered through non-AI means.² However, no AI-discovered therapy has thus far made it through phase 3 trials.

“The question of whether AI can impart meaningful, sustained disruption to drug development has remained unanswered,” they wrote.¹

Patients in the 60 mg arm had a mean increase in FVC of 98.4 ml while the placebo group had a drop in FVC of 20.3 ml. | Image credit: Angelov - stock.adobe.com

The new report outlined their efforts to evaluate the utility of an AI-generated novel target for IPF. The group’s AI discovery tools identified Traf2 and Nck-interacting kinase (TNIK) as important regulators of IPF pathology. The authors noted that the target plays a role in multiple profibrotic and proinflammatory cellular processes. They then used generative AI tools to develop a small-molecule TNIK inhibitor, rentosertib. In phase 1 data published last year, the therapy was found to be safe and tolerable and to demonstrate anti-fibrotic and anti-inflammatory effects.³

“This was also, importantly, the first reported instance of AI platform-enabled discovery of both a disease-associated target and a compound for that target,” they wrote. They added that it took just 18 months to identify the drug candidate and just 30 months from the start of target discovery to complete phase 0/1 testing.

The new phase 2a study was a multicenter, double-blind, randomized, placebo-controlled trial to examine the safety and tolerability of the therapy, as well as its pharmacokinetics and impact on forced vital capacity (FVC) and self-reported quality of life metrics.

The investigators enrolled 71 patients in the trial and assigned them to receive placebo (n = 17) or rentosertib (n = 54). Those who received the therapy were randomized into one of 3 dosage arms, each with 18 patients. The first arm received one 30 mg dose per day; the second group received two 30 mg doses per day, and the third group received one 60 mg dose daily.

After 12 weeks of treatment, most patients had at least 1 treatment-emergent adverse event (TEAE), but the rates were similar across all of the arms, including the placebo arm. For instance, the 30 mg once-daily arm had a TEAE rate of 72.2%, compared with 70.6% for the placebo group. The other two arms both had 83.3% TEAE rates.

The most common TEAEs were hypokalemia, abnormal hepatic function, diarrhea, and alanine aminotransferase increase. Serious TEAEs occurred in a minority of all of the study arms. Three patients in the placebo group had serious adverse events, compared to 2 in the 30-mg once-daily group, 4 in the 30-mg twice-daily group, and 7 in the 60-mg once-daily group. The most common adverse events that led to treatment discontinuation were liver toxicity and diarrhea.

Patients in the 60 mg arm had a mean increase in FVC of 98.4 ml (95% CI, 10.9-185.9). The placebo group had a drop in FVC of 20.3 ml (95% CI, –116.1 to 75.6).

In terms of quality of life, the data were “largely inconclusive” due to large variance between the different treatment arms, the authors explained, though they noted that patient-reported coughing-related metrics improved in the 60 mg arm.

The investigators said they are encouraged by the results, but more work is needed to validate the target and therapy.

“Longer phase 2 or 3 trials studying rentosertib in a larger, heterogeneous sample of patients with IPF from around the globe are warranted to further evaluate the efficacy and safety of rentosertib treatment for IPF,” they concluded.

References

1. Xu Z, Ren F, Wang P, et al. A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial. Nat Med. Published online June 3, 2025. doi:10.1038/s41591-025-03743-2

2. KpJayatunga M, Ayers M, Bruens L, Jayanth D, Meier C. How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons. Drug Discov Today. 2024;29(6):104009. doi:10.1016/j.drudis.2024.104009

3. Ren F, Aliper A, Chen J, et al. A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models. Nat Biotechnol. 2025;43(1):63-75. doi:10.1038/s41587-024-02143-0