Fresh NIH funding backs an iPSC-and-AI push to find drugs for Duchenne cardiomyopathy without relying on mouse models that rarely translate.
By the numbers, the translational gap is expensive. Roughly nine in ten drug candidates that succeed in animal models still fail once they reach human trials, and cardiology has one of the worst records of all. Mouse hearts beat around 600 times a minute; human hearts beat around 60. That single mismatch in electrophysiology has been blamed for a long list of cardiac programs that looked promising in rodents and collapsed in the clinic – a frustrating bottleneck for geroscience, where cardiac fibrosis and cellular exhaustion in conditions like DMD directly mirror the hallmark pathologies of intrinsic cardiac aging.
Greenstone Biosciences, a Palo Alto biotech built around a biobank of more than 2,500 donor-derived induced pluripotent stem cell (iPSC) lines, has just received fresh federal backing to test whether skipping the animal step altogether can produce better drugs, faster [1].
The National Heart, Lung, and Blood Institute has awarded Greenstone a Catalyze R61 grant to fund a discovery program for myocardial fibrosis and dilated cardiomyopathy in Duchenne muscular dystrophy (DMD), a rare genetic disorder affecting around one in every 3,500 baby boys worldwide.
Heart failure, driven by progressive scarring of cardiac tissue, is the leading cause of death in DMD patients. Corticosteroids can slow the decline, but more than a quarter of patients cannot tolerate them or do not respond, and no approved therapy directly targets the underlying fibrosis. That gap is precisely where Greenstone intends to work.
Human cells, not mouse proxies
Instead of screening compounds in animal hearts, Greenstone will run its discovery program on cardiomyocytes differentiated from its own iPSC biobank, layering in unbiased proteomics, computational drug screening, and generative AI to identify and validate candidates.
The company’s pitch is straightforward: a drug that behaves well on human heart cells with the actual genetic background of DMD patients should be a safer bet in the clinic than one that merely behaved well in a rodent.
“DMD cardiomyopathy has gone unaddressed for too long, in part because the tools to model it faithfully in the lab have not existed,” said Dr Joseph C Wu, Greenstone cofounder and Director of the Stanford Cardiovascular Institute.
The R61 award funds target identification and early candidate discovery; a successful phase would trigger a follow-on R33 award to fund synthesis, characterization and in vivo testing ahead of IND-enabling studies.
Where AI earns its keep
Generative AI has become a fixture of drug-discovery marketing, but its value depends heavily on what it is trained on. A model trained on synthetic or animal-derived data can only ever be as human-relevant as its inputs allow.
Greenstone’s pitch is different: its algorithms are trained against outputs from a biobank the company describes as one of the largest human iPSC collections in the world, including an unusually deep set of rare and orphan disease cell lines.
For a condition like DMD cardiomyopathy, where patient numbers are small and traditional trial-and-error drug development is slow and expensive, that combination could matter more than in larger disease categories. Whether AI trained on patient-derived cellular data can meaningfully shorten discovery timelines for orphan diseases is still an open empirical question, but it is exactly the kind of question this program is positioned to answer.
The timing is not incidental. The FDA’s 2025 Roadmap to Reduce Animal Testing and an April 2026 progress report have made the agency’s ISTAND pilot program permanent, created a New Approach Methodologies (NAMs) Acceptability Database, and cleared the FDA’s first AI-based drug development tool for use in clinical trials.
Congress, meanwhile, has advanced the FDA Modernization Act 3.0, which would formally align FDA regulations with the nonclinical testing authority it was granted in 2022; the bill passed the House on 20 July 2026, and awaits a Senate vote.
Together, these moves represent some of the strongest regulatory signals yet that human-relevant models – rather than animal testing by default – could become an accepted, even preferred, pathway for early drug development.
Greenstone is not promising to eliminate late-stage clinical failure outright, and no single grant will settle the animal-versus-human-model debate. But a validated candidate emerging from this program would be a meaningful proof point: that a company can go from iPSC biobank to AI-assisted candidate to a clinic-ready molecule for a disease with no approved fibrosis therapy, without ever putting a compound through a mouse first. For a field still arguing over how much animal data regulators should require, that would carry weight well beyond DMD.
