BIGGER PICTURE Better care
An AI-designed lung drug reached patients. What did the trial find?
Rentosertib links two difficult searches: which biological target might change lung fibrosis, and what molecule could act on it? The AI-assisted answers reached a randomized trial in people with the disease.

- Trial size
- 71 people with pulmonary fibrosis
- Treatment period
- 12 weeks
- Highest-dose lung-function change
- +98.4 mL versus −20.3 mL with placebo
How the idea works
Choose a fibrosis target
Models prioritise TNIK from disease-related biological evidence.
Design an inhibitor
Generative chemistry proposes a molecule to inhibit TNIK.
Test against placebo
Randomized dosing tests safety and lung-function signals in patients.
A disease that progressively stiffens the lungs
Idiopathic pulmonary fibrosis involves progressive scarring of lung tissue. The research programme used AI to prioritise TNIK, a protein involved in programmes linked to fibrosis and inflammation. A second design task produced a small molecule intended to inhibit that target.
Sources: [1]
Predictions still have to meet biology
The 2025 phase 2a study enrolled 71 patients in China. Participants were randomly assigned to three rentosertib regimens or placebo for twelve weeks, with patients and investigators blinded to allocation. The main outcome concerned adverse events; lung-function measures were secondary outcomes.
Sources: [1]
A change in the volume patients could exhale
In the highest-dose group, forced vital capacity increased by a mean 98.4 millilitres, versus a decline of 20.3 millilitres with placebo. This measure records how much air someone can forcefully exhale. The encouraging difference warrants a larger test; it does not establish longer survival or reversal of disease.
Sources: [1]
The important achievement is testability
Adverse events occurred in every group; liver toxicity and diarrhoea were among reasons for discontinuation. Small groups and short follow-up limit conclusions. What makes this case valuable is the complete chain from a machine-suggested biological hypothesis to a molecule and a controlled clinical experiment.
Sources: [1]