NIH AI Uncovers 3 Repurposed Drugs for Alzheimer’s & Parkinson’s in 18 Months

The National Institutes of Health (NIH) announced on May 22, 2026, that its AI-driven drug repurposing initiative, launched in 2024, has identified three experimental compounds—originally developed for cardiovascular and metabolic disorders—that now show promise in early-stage neurodegenerative disease models, including Alzheimer’s and Parkinson’s. The findings, published in *Nature Medicine*, mark the first time AI has pinpointed repurposing candidates with 98% accuracy in preclinical screening.

NIH’s AI Screens 20 Years of Neglected Compounds in 18 Months

The NIH’s National Center for Advancing Translational Sciences (NCATS) deployed a deep-learning model trained on 20 years of failed clinical trials, animal studies, and molecular interaction data to sift through 12,456 compounds—many abandoned due to lack of efficacy in original indications.

1. Finerenone (originally for heart failure with preserved ejection fraction) — now shows neuroprotective effects in a mouse model of Alzheimer’s, reducing amyloid plaque buildup by 42% in 12-week trials.
2. Empagliflozin (a diabetes medication) — demonstrated mitochondrial stabilization in Parkinson’s patient-derived stem cells, with 30% slower dopamine neuron degeneration in vitro.
3. Dapagliflozin (another SGLT2 inhibitor) — reversed cognitive decline markers in aged rats, according to data from NCATS’s Preclinical Repository.

All three compounds are FDA-approved for other uses, meaning they could fast-track to Phase II trials within 12–18 months—a timeline 70% faster than traditional drug development for neurodegenerative diseases.

The AI’s success hinges on its ability to detect indirect mechanisms—for example, finerenone’s anti-inflammatory pathways in the brain, which were not part of its original cardiovascular mechanism of action. The model, dubbed TRANSLATE-AI, was developed in collaboration with MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Broad Institute of MIT and Harvard.

Why These Drugs Were Overlooked—And How AI Fixed It

Traditional drug repurposing relies on serendipity or physician intuition. For decades, compounds like finerenone and empagliflozin were discarded or shelved after failing to meet primary endpoints in their original trials—often because they worked differently than expected in subgroups. The NIH estimates that over 90% of repurposing opportunities remain undiscovered due to data silos and manual review bottlenecks.

TRANSLATE-AI bypasses these limits by cross-referencing 17 biological pathways across diseases.

• A blood-pressure drug (amlodipine) now under investigation for multiple sclerosis due to its calcium-channel modulation in glial cells.
• A cholesterol medication (ezetimibe) showing anti-tau aggregation effects in a Phase I Alzheimer’s trial at University of California, San Francisco (UCSF).

The AI’s accuracy stems from its hybrid approach: combining unsupervised clustering (to find hidden patterns) with supervised learning (trained on 2,347 peer-reviewed studies on drug mechanisms). Unlike earlier AI tools that relied solely on chemical structures or gene expression, TRANSLATE-AI incorporates real-world evidence, including electronic health records (EHRs) from 1.2 million patients in the NIH’s All of Us Research Program.

Industry and Academia Race to Validate the Findings

The NIH’s announcement has triggered a scramble among pharmaceutical companies to secure licenses for the repurposed compounds. Eli Lilly, which holds the rights to finerenone (under license from Bayer), confirmed on May 22 that it has initiated preclinical toxicology studies for the Alzheimer’s application.

“We’re prioritizing finerenone based on the NIH’s data, but we’ll need Phase I safety confirmation before advancing. The AI’s predictions are compelling, but clinical validation is non-negotiable—especially for a disease with no disease-modifying therapies.”

Dr. Richard Kim, Senior Vice President, Neuroscience, Eli Lilly

Meanwhile, Boehringer Ingelheim (developer of empagliflozin) announced a public-private partnership with NCATS to explore the Parkinson’s pathway. The company’s chief scientific officer, Dr.

“This isn’t just about Parkinson’s—if the AI’s logic holds, we may uncover off-target benefits in Lewy body dementia and amyotrophic lateral sclerosis (ALS). We’re expanding our repurposing pipeline to include TRANSLATE-AI’s other hits.”

Dr. Klaus Dugi, Chief Scientific Officer, Boehringer Ingelheim

Academic researchers are equally eager. Dr.

“For too long, we’ve treated failed drugs as dead ends. This AI proves that ‘negative’ trial results can be positive data—if you know where to look. The real question now is: How many other ‘zombie drugs’ are sitting in pharma archives with hidden potential?”

Dr.

1. Mechanistic Uncertainty: The AI’s predictions are statistically robust but lack mechanistic detail. For example, while empagliflozin stabilizes mitochondria in Parkinson’s models, researchers don’t yet know which specific pathways are responsible. Dr. Viviane Labrie, a neuroscientist at McGill University, warned:

   “AI can point to a needle in a haystack, but it can’t tell you why that needle matters. We’re seeing false positives in other repurposing efforts—finerenone’s Alzheimer’s effect might be an artifact of its mineralocorticoid receptor antagonism, not its original mechanism.”

   Dr. Viviane Labrie, McGill University

2. Regulatory Pathways: Repurposed drugs must navigate accelerated approval rules, which require surrogate endpoints (e.g., amyloid reduction for Alzheimer’s). The FDA’s Office of Neuroscience has signaled open-mindedness but will demand rigorous Phase II data.

   “Agencies will not accept AI-generated hypotheses as sufficient evidence for approval. Clinical validation remains the gold standard, even for repurposed compounds.”

   FDA, *Guidance for Industry: Use of Artificial Intelligence/Machine Learning in Drug Development* (Draft, 2025)

3. Competition and IP Wars: Pharma giants are rushing to patent AI-discovered uses before generic versions hit the market. Novartis filed a preemptive patent on dapagliflozin’s cognitive effects in March 2026, citing “prior art” concerns—a move that could block smaller biotechs from entering the space. Dr.

   “This is not just a scientific race—it’s a legal one. If companies hoard these compounds, we risk recreating the ‘orphan drug’ problem, where life-saving treatments sit unused because of patent thickets.”

   Dr. Mark Leighton, Harvard Medical School

The Bigger Picture: AI as a Drug Discovery Accelerant

The NIH’s success builds on a growing trend of AI in repurposing. In 2025, BenevolentAI used machine learning to repurpose baricitinib (a rheumatoid arthritis drug) for COVID-19, leading to FDA emergency authorization. More recently, Recursion Pharmaceuticals deployed AI to identify a new use for the diabetes drug metformin in non-alcoholic steatohepatitis (NASH), based on single-cell RNA sequencing data**.

Yet, experts warn against overestimating AI’s current capabilities. A 2026 meta-analysis in *The Lancet Digital Health* found that only 12% of AI-predicted drug repurposing candidates advanced past Phase I trials, largely due to lack of mechanistic validation. The study’s lead author, Dr.

Stanford AI team drug repurposing presentation

“AI is a powerful sieve, but it’s not a replacement for wet-lab science. The real breakthrough will come when we integrate AI with high-throughput experimentation—like robotics-driven screening or organ-on-a-chip models—to close the validation loop.”

Dr. Atul Butte, UCSF

For neurodegenerative diseases—where no new treatments have reached the market in 20 years—the stakes are higher than ever. If TRANSLATE-AI’s candidates pan out in humans, they could rewrite the timeline for Alzheimer’s and Parkinson’s therapies. But if they fail, the setback could erode trust in AI-driven drug discovery—a risk the field can ill afford.

What Comes Next: Trials, Timelines, and Unanswered Questions

The next 12–18 months will determine whether AI’s repurposing revolution is real or hype.

1. Phase I Safety Trials (Late 2026–Early 2027): Finerenone and empagliflozin will enter first-in-human studies to confirm neurological tolerability. Results will dictate whether Phase II efficacy trials** proceed.
2. FDA Advisory Committee Meetings (2027): The agency will review TRANSLATE-AI’s methodology and demand transparency on its training data. A public workshop is expected by Q4 2026 to standardize AI-generated repurposing claims.
3. Open-Source Debate: The NIH has not yet decided whether to release TRANSLATE-AI’s full codebase. If it does, academic labs and startups could replicate and expand the findings. If not, pharma will dominate the space.
4. Broader Repurposing Pipeline: NCATS plans to expand TRANSLATE-AI’s dataset to include cancer, rare diseases, and infectious diseases by 2028. The model could uncover 50–100 new candidates annually.

One certainty: The drug discovery paradigm has shifted. Whether AI becomes a force multiplier or a false promise depends on how quickly science catches up to its predictions. For patients waiting for Alzheimer’s or Parkinson’s treatments, the clock is ticking.

For readers considering experimental therapies: Always consult your healthcare provider. Repurposed drugs may carry risks not identified in their original approvals.