A new Alzheimer's drug usually takes more than a decade and costs tens of billions of dollars to develop. A government agency asked a simpler question: what if a treatment already exists and nobody noticed?
NCATS, the National Center for Advancing Translational Sciences, is the part of the NIH that turns lab discoveries into actual patient treatments.It built an open-source knowledge graph-based system called the Biomedical Data Translator to scan thousands of approved drugs and research compounds for uses nobody originally intended.The system pulls from over 250 knowledge sources and roughly 94 million connections mined from published medical research, hunting for patterns that suggest an existing medication might work against a disease it was never designed for.
For Alzheimer's, that approach is producing real candidates.
Bumetanide is a diuretic. Doctors have prescribed it for decades to treat swelling and high blood pressure.Now it's in a Phase IIa clinical trial at Stanford for Alzheimer's disease. Computational analysis found the drug appeared to reverse Alzheimer's-related gene expression patterns, and health record data showed patients already taking bumetanide had significantly lower Alzheimer's diagnosis rates.
XPro1595, an anti-inflammatory compound,missed its primary endpoint in the overall population in a Phase 2 trial, but demonstrated directionally consistent cognitive benefits in a prespecified subgroup of Alzheimer's patients with high inflammatory burden, with no occurrences of amyloid-related imaging abnormalities (ARIA).
The economics make the approach hard to ignore.Traditional drug development takes 3 to 20 years and burns through tens of billions. Starting with a drug that's already been tested in thousands of patients skips the most expensive part of that timeline. NCATS's library alone contains roughly 10,000 compounds for the AI to sift through, including about 3,700 that are already approved or under investigation.
Jeffrey Cummings, a neuroscientist at the University of Nevada, Las Vegas,said it directly: "Alzheimer's is no longer an untreatable disease."
That optimism comes with a very long asterisk. Most anti-inflammatory drugs tested for Alzheimer's have failed in trials,and some have actually made cognition worse. Semaglutide, the blockbuster weight-loss drug many hoped could double as an Alzheimer's treatment,struck out in two Phase 3 trials00459-9). Evan Feinberg, founder of Genesis Molecular AI,stated in the official partnership announcement: "High-quality proprietary data is among the most valuable inputs for advancing molecular AI, and our expanded collaboration will enable both companies and patients to benefit from an industrial-scale flywheel..."
NCATS Director Joni Rutter sees this as just the beginning.The agency director's 'Audacious Goals' presentation targets a five-fold increase in the number of diseases with available treatments, aiming for "more treatments for all people more quickly."
AI isn't curing Alzheimer's. What it's doing is sorting through the thousands of drugs we've already proven safe and asking whether any of them do something we never realized. Most won't. But the ones that do could reach patients years faster and for a fraction of what a new drug would cost. The real question isn't whether AI can surface promising candidates, because it clearly can. It's whether those candidates can survive the clinical trials that have destroyed nearly everything else thrown at this disease.
