The Robot That Designed a Lung Drug (and It Actually Worked)

Patients on the 60 mg dose went the other direction: their FVC increased by about 98 mL. And in the subgroup of patients who weren't already taking standard antifibrotic drugs, the improvement was even more striking, at roughly 188 mL.

For context, existing IPF drugs are considered successful when they slow a decline. This drug showed a gain. In a disease where lungs only go one direction, that's like watching a car roll uphill.

The Robot Behind the Curtain

What makes rentosertib different from every other IPF drug in development isn't just the clinical data. It's how it was born.

Insilico's AI platform, called Pharma.AI, handled two jobs that usually take humans years and hundreds of millions of dollars. First, its target discovery engine (PandaOmics) crunched mountains of genomic, clinical, and scientific literature data to identify a protein called TNIK as a promising drug target in fibrosis. TNIK sits at the crossroads of several pathways that drive scar tissue formation, almost like a traffic cop directing multiple lanes of fibrotic signaling.

Second, the company's generative chemistry engine (Chemistry42) designed a molecule to block TNIK. Running over 30 AI models in parallel, it optimized for potency, selectivity, and drug-like properties.

The whole process, from picking the target to nominating a clinical candidate, took under 18 months and cost roughly $2.6 million. The traditional version of that journey? Typically four to six years and tens of millions of dollars. Getting from target to Phase 1 took under 30 months total.

Rentosertib is now the first drug where both the target and the molecule were generated by AI and then validated in a randomized, placebo-controlled human trial.

Not All Roses: The Liver Question

Before anyone gets too excited, there's a yellow flag on the field. Seven patients dropped out due to liver-related issues, and four of those were also taking nintedanib, which has its own liver baggage. In the 60 mg group, only 12 of 18 patients (67%) completed the full 12 weeks, compared to 88% on placebo.

Insilico's CEO Alex Zhavoronkov has described the liver enzyme elevations as transient and reversible, mostly appearing in patients on background therapy. That's plausible, but it's a problem that needs a clean answer before larger trials can move forward. Drug-drug interactions with existing IPF medications will be a critical puzzle to solve.

Why This Matters Beyond IPF

The IPF results are genuinely interesting on their own. But the bigger story is what this means for AI-driven drug discovery as a field.

For years, the sector has been long on promises and short on clinical proof. Exscientia was the first to get an AI-designed molecule into human trials. Recursion has built the broadest clinical pipeline, with multiple Phase 2 programs. But Insilico is the first to show positive efficacy data from a trial where AI handled the entire discovery process, target and molecule alike.

That distinction matters to investors, pharma partners, and skeptics. It's one thing to say AI can speed up drug discovery. It's another to point to a peer-reviewed Nature Medicine paper showing it produced a drug that outperformed placebo in a notoriously difficult disease.

Is this proof that AI will revolutionize all of pharma? No. One Phase 2a win in 71 patients doesn't rewrite the industry playbook. The confidence intervals are wide, the trial was short, and plenty of drugs that look good in Phase 2 collapse in Phase 3.

But it's the strongest evidence yet that the AI drug discovery thesis isn't just hype. The robot designed a drug. The drug worked in humans. Now Insilico has to prove it works in a bigger trial, for longer, without melting anyone's liver.

What Comes Next

Insilico is gearing up for larger trials. The company has also received IND clearance in China for an inhaled version of rentosertib, which could theoretically deliver the drug directly to the lungs and sidestep some of those systemic side effects.

For the IPF community, this is a cautiously hopeful moment. For the AI drug discovery world, it's the first real receipt. Not a press release, not a preclinical poster; actual human data showing that an AI-born molecule can do something meaningful in a disease that desperately needs new options.

The bar is still high. But for the first time, an AI-designed drug just cleared it.

Trodelvy Just Jumped the Line in Breast Cancer Treatment

The FDA just moved Gilead's Trodelvy from last-resort to first-choice in triple-negative breast cancer, the deadliest breast cancer subtype. It's a seismic shift for patients, a multi-billion-dollar opportunity for Gilead, and a surprisingly friendly outcome for Merck's Keytruda.