An AI-designed drug has shown meaningful improvements in patients with a fatal lung disease, according to results published in Nature Medicine. This marks the first time a drug discovered and designed entirely using generative AI has demonstrated clinical efficacy in a randomized trial.
The Study
Insilico Medicine ran a 12-week, double-blind trial called GENESIS-IPF across 22 sites in China. They enrolled 71 patients with idiopathic pulmonary fibrosis (IPF), a progressive lung disease that typically kills patients within 3-5 years of diagnosis. Patients were randomly assigned to receive either placebo or one of three doses of rentosertib (formerly ISM001-055): 30 mg once daily, 30 mg twice daily, or 60 mg once daily.
The results showed a clear dose-response pattern. Patients taking the highest dose (60 mg once daily) gained an average of 98.4 mL in forced vital capacity (FVC), a measure of lung function. Meanwhile, patients on placebo lost an average of 20.3 mL. That’s a difference of nearly 120 mL in lung capacity over just 12 weeks.
For context, existing IPF drugs like pirfenidone slow decline but don’t reverse it. A drug that actually improves lung function would be meaningful for patients who currently face a one-way trajectory toward respiratory failure.
How AI Built This Drug
What makes this trial unusual isn’t just the results - it’s how the drug came to exist. Insilico used its generative AI platform to identify TNIK (Traf2- and Nck-interacting kinase) as a disease target, then designed the actual drug molecule to hit that target. Both steps - finding what to target and creating a drug to target it - were AI-driven.
The company claims this cut their discovery timeline substantially compared to traditional methods, though precise comparisons are difficult because different drugs have different complexity. What’s verifiable is that they went from AI-generated molecule to Phase 2a results faster than most pharmaceutical programs manage.
The Fine Print
Several caveats apply. First, the sample size was small - roughly 18 patients per group. The confidence intervals are wide, and some groups showed inconsistent results. The 30 mg twice-daily group, despite receiving more drug than the 30 mg once-daily group, showed smaller improvements.
Second, 12 weeks is short for a chronic disease like IPF. Whether these improvements persist or compound over longer periods remains unknown. The company will need larger, longer trials to prove this isn’t a temporary effect.
Third, this was conducted entirely in China. While that doesn’t invalidate the results, regulatory approval in the US and Europe will require additional trials in those populations.
Safety data was encouraging - adverse event rates were similar across all groups, most events were mild or moderate, and serious adverse events were rare.
What This Means for AI Drug Discovery
The pharmaceutical industry has been promising AI-discovered drugs for years. Most of those promises have been aspirational - algorithms that find targets, predict structures, or optimize molecules, but with minimal clinical validation.
This trial changes that. Rentosertib is the first AI-designed drug to show efficacy in a randomized, controlled human trial for an AI-discovered target. It’s not a cure, and it needs larger trials, but it’s evidence that the full AI drug discovery pipeline can produce something that actually works in patients.
The bigger question is whether this is repeatable. Insilico has over 30 other AI-designed programs in development. If several of those show similar results, AI drug discovery moves from “promising” to “proven.” If they don’t, rentosertib might be an outlier rather than a template.
For now, the results are what they are: a small trial with encouraging signals. IPF patients have reason for cautious hope. The rest of us have reason to take AI drug discovery more seriously than we did last week.