Imagine waking up every morning and the first thing you feel is your own lungs - not with a deep, refreshing breath, but with a quiet, creeping tightness. A cough that won't quit. A body that's slowly forgetting how to breathe. That's the daily reality for the roughly 3 million people worldwide living with idiopathic pulmonary fibrosis, a rare lung disease that scars the lungs from the inside. Doctors call it progressive, incurable, and ultimately fatal. Most patients are given 3 to 5 years from diagnosis.
For decades, medicine had no real answer. Until, in 2024, an artificial intelligence did what an entire generation of researchers could not - it designed a drug that actually reversed the damage. This is not science fiction. The results were published in Nature Medicine, one of the world's most respected scientific journals, in June 2025. And they are extraordinary.
A Disease With No Name, No Cause, No Cure
"Idiopathic" is a medical word that essentially means we don't know why this happens. In IPF, scar tissue - called fibrosis - slowly forms across the tiny air sacs inside the lungs. These sacs, of which your lungs contain roughly 300 million, are responsible for transferring oxygen into your blood every time you breathe. When they scar over, they stiffen. The walls thicken. Breathing becomes like trying to inflate a balloon made of rubber bands.
Patients describe climbing a single flight of stairs as exhausting. Simple things - talking on the phone, laughing, eating a meal - can trigger coughing fits that last minutes. Over time, many patients become dependent on oxygen tanks just to move around their own homes. And because IPF shares early symptoms (dry cough, fatigue, breathlessness) with other conditions like asthma or heart disease, it is frequently misdiagnosed for months, sometimes years. By the time a confirmed diagnosis arrives, the disease has already been quietly winning.
The three FDA-approved treatments available today - nintedanib, pirfenidone, and nerandomilast - can slow the scarring. But they cannot stop it. They cannot reverse it.
No drug, until now, had ever made an IPF patient's lungs measurably better.
Enter the Algorithm That Changed Everything
In Cambridge, Massachusetts, a biotech company called Insilico Medicine was asking a different question. Instead of searching for a new version of existing drugs, their AI platform - called Pharma.AI - was tasked with finding an entirely new biological target: something in the body that was contributing to fibrosis that nobody had thought to target before.
The AI surfaced a protein called TNIK (Traf2- and NCK-interacting kinase). It's not a protein that shows up in traditional IPF research. But the AI had scanned biological databases, genetic patterns, and molecular interaction maps at a scale no human team could match - and it identified TNIK as a potential driver of the scarring process. Then it went further. It didn't just find the target. It designed the molecule to hit it.
AI, as an advanced technology, is already playing a crucial role in many aspects of medical practice, including drug discovery and clinical research, and we expect to see the real clinical benefits it brings to patients.
- Dr. Zuojun Xu, Professor at Peking Union Medical College & Lead Investigator, Phase IIa Trial
That molecule became Rentosertib (also known as ISM001-055). And the speed at which it was created is almost as remarkable as the results.
46 Days. What Normally Takes Over a Year
This is the part that should make anyone stop and re-read it. From concept to a validated compound ready for further development, Insilico's AI completed the design and validation of Rentosertib in just 46 days. The same process - finding a promising molecule, testing it computationally against biological targets, refining its properties - typically takes over a year using conventional methods. Sometimes several years.
To understand why that speed matters, consider the economics of drug discovery. Bringing a single drug from idea to pharmacy shelf takes, on average, 10 to 15 years and over $2 billion. Most drugs fail - roughly 90% of candidates never make it to patients. When AI compresses the early phases of that timeline, it doesn't just save time. It saves the resources to pursue more diseases, especially rare ones that wouldn't otherwise be commercially viable to research.
The Trial That Proved It Wasn't Just Hype
Calling something "AI-designed" is easy. Proving it works in a real human body is another matter entirely. That's where the Phase IIa trial - published in Nature Medicine in June 2025 - becomes critical. This was a double-blind, placebo-controlled trial. The gold standard of medical evidence. Neither the patients nor the doctors administering treatment knew who was receiving Rentosertib and who was receiving a placebo.
71 IPF patients across 21 clinical sites in China were enrolled. They were randomly assigned to different doses of Rentosertib, or a placebo, for 12 weeks. The key measurement was forced vital capacity (FVC) - essentially, how much air a patient can push out of their lungs in one breath. In IPF, this number usually declines over time. That's the disease winning.
In the placebo group, FVC declined by an average of 20.3 mL over 12 weeks. That's the disease doing what IPF does: slowly suffocating the lungs.
In the group receiving the highest dose of Rentosertib (60 mg), FVC improved by 98.4 mL. Not slowed. Not plateaued. Improved. That's a swing of nearly 119 mL between the drug group and the placebo - a gap that, in the context of this disease, is clinically enormous. Patients also reported meaningful improvements in quality of life scores on the Leicester Cough Questionnaire. They were coughing less. Breathing more comfortably. Living a little more like themselves again.
It not only reflects ISM001-055's potential to slow disease progression but also suggests its capability to stop or even reverse it.
- Dr. Zuojun Xu, Lead Investigator, Nature Medicine, June 2025
This Isn't a One-Off. It's a Pattern
Rentosertib is the most compelling proof point, but it isn't an isolated story. Across the pharmaceutical world, AI is beginning to prove its worth in clinical settings - not just in theory, but in results.
In 2020, a UK company called Exscientia collaborated with Japanese pharmaceutical giant Sumitomo Dainippon Pharma to create the world's first AI-designed drug to enter human trials. The drug, DSP-1181, was developed for obsessive-compulsive disorder (OCD) - a condition affecting millions who experience relentless, intrusive thoughts and compulsive behaviors that can make ordinary life feel impossible. Traditional drug discovery for a molecule like this takes 4 to 5 years just to reach the trial stage. Exscientia's AI platform, Centaur Chemist, completed the exploratory research in under 12 months - screening 350 candidate compounds instead of the typical 2,500.
And then there's the data that might be most surprising of all. Research published in 2024 found that AI-discovered drugs have an 80–90% success rate in Phase I human trials - compared to 40–65% for conventionally discovered drugs. AI isn't just faster. It is, increasingly, smarter about which molecules will actually work.
What This Means for You
You may never have heard of IPF before today. But the significance of what just happened isn't limited to one rare lung disease. What Insilico proved - and what Exscientia demonstrated before them - is that the entire process of finding new medicines can be accelerated, made more precise, and extended to conditions that have historically been ignored because the economics of traditional drug discovery made them too expensive to pursue.
Rare diseases. Drug-resistant infections. Alzheimer's. Conditions where human researchers have been chasing dead ends for decades. AI can approach these problems without fatigue, without bias toward what's worked before, and without being limited to the number of compounds a laboratory can physically screen in a year. The AlphaFold AI system - which solved the 50-year-old problem of predicting protein structure and won its creators the 2024 Nobel Prize in Chemistry - is now the foundation on which many of these drug discovery AIs build their understanding of molecular biology. Every new AI-designed drug candidate benefits from it.
AI is not replacing doctors, researchers, or the rigour of clinical trials. Rentosertib still had to prove itself in a real trial, on real patients, with real medical oversight. But it got to that trial faster, targeting something no human had thought to target - and it worked.
The Breath That Tells the Whole Story
Somewhere in China, among the 71 patients in that Phase IIa trial, there are people who enrolled knowing their lungs were deteriorating. Who sat in a clinic for 12 weeks, not knowing whether they were receiving a drug or a sugar pill. And at the end of those 12 weeks, the patients who received the highest dose of Rentosertib breathed - measurably, scientifically, provably - better than when they started.
That's what medicine is supposed to do. And for the first time in the history of IPF, it's AI that made it happen.
The drug is now advancing to larger Phase IIb trials. If those go well, Phase III. And if Phase III confirms what 2024 showed - then one day, not too far from now, an IPF patient somewhere will be prescribed a drug that an algorithm designed in 46 days. A drug that exists because a machine looked at biology differently than any human had before.
Sources
- A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis - nature.com
- AI drug enters human clinical trials - InformationAge