An AI system scanned over 2 million biological samples, picked an old heart drug called ouabain, and scientists showed it really improved aging markers in mice.
An AI system called ClockBase Agent was recently put to a tough test: could it search huge piles of old biology data, pick out a drug that might slow aging, and then be proven right in real animals?
The answer is yes, at least once so far. The AI flagged ouabain, a little-known drug, as a top candidate for slowing aging. When researchers tried it in 20-month-old mice (roughly senior age for a mouse), the animals became healthier, less frail, and showed better heart and brain function compared to untreated mice.
Key takeaways
- ClockBase Agent, an AI system, scanned over 2 million human and mouse samples to spot shifts in biological age.
- It tested 43,529 interventions and flagged 5,756 as likely to change how fast we age.
- One standout was ouabain, a little-known drug in longevity circles.
- In old mice treated with ouabain for about 3 months, the animals were less frail, had healthier hearts, better fur, and improved cognition compared to controls.
- The AI also confirmed classics like rapamycin and metformin, but most drugs it found actually sped up aging, not slowed it.
- The research is still a preprint, not peer-reviewed, and ouabain can be dangerous in humans, so this is not something to try on yourself.
What did the AI actually do?
🤖 ClockBase Agent was built to answer one big question: What makes biological age go up or down?
To do this, the system looked at more than 2 million samples from humans and mice. These samples included gene expression data (which genes are turned on or off) and epigenetic data (chemical marks on DNA that change with age). All of this came from public databases like the Gene Expression Omnibus (GEO), which holds decades of experiments that were mostly not designed for aging research.
On top of that, the AI used 40 different aging clocks. These clocks are algorithms that look at patterns in the data and estimate biological age, which is more about health and wear-and-tear than calendar age. Some of these clocks, like AdaptAge, CausAge, and DamAge, try to focus on cause and damage, not just correlation.
👉 In simple terms:
- The AI asked: “When this drug, gene, or condition is present, do the clocks say the tissue looks younger or older than expected?”
- It then used a large language model (LLM) to form hypotheses, checked the original data again, and read the related scientific papers to see if the idea made sense.
This is different from older AI approaches that only tried to match drugs to age-related patterns without really checking context. Here, the AI behaved more like a junior scientist, making suggestions and then double-checking itself.
What scientists found inside the data
🧪 Using this setup, ClockBase Agent analyzed 43,529 interventions. These were not just drugs, but also:
- Gene knockouts and overexpression
- Diseases and infections
- Environmental stressors
- Lifestyle-like factors, such as mechanical overload that may resemble resistance training
From these, the AI decided 5,756 interventions likely had a real impact on biological aging.
Some examples:
- It agreed with human-built databases that rapamycin and metformin lower biological age.
- It identified environmental patterns where mechanical overload plus senolytic treatment (drugs that clear senescent cells) appeared to lower biological age.
- It saw that hypoxia, ischemia–reperfusion injury (like what happens around heart attacks), some viral infections, and intense light exposure in embryos seemed to speed up aging.
The AI’s findings overlapped strongly with GeneAge and other curated databases, which is a good sign. But it also surfaced many signals that were not yet in aging drug databases, meaning they might have been hiding in plain sight in older experiments.
Ouabain: an old drug with a new role?
One of the most striking AI picks was ouabain.
💊 Ouabain is a cardiac glycoside, a type of drug that affects the sodium–potassium pump in cells and has long been used in heart-related research. It is also considered a senolytic, meaning it might help clear out senescent cells, which are old, damaged cells that stop dividing but stay in the body and secrete harmful signals.
ClockBase Agent reported that, across multiple datasets, ouabain strongly reduced biological age according to several aging clocks. Many researchers had not thought of ouabain as a top longevity drug candidate before.
So the team did something crucial: They tested ouabain in real animals.
The mouse test: Did the AI prediction hold up?
🐭 Researchers took 20-month-old C57BL/6 mice (already old by mouse standards) and gave one group intermittent ouabain for about 3 months, while another group served as controls. The dosing protocol matched the experiments the AI had used in its analysis.
By the end of the study, the ouabain-treated mice were:
- Less frail, based on standard frailty scores
- Showing better cognitive performance on behavior tests
- Displaying healthier fur and overall condition
- Having better heart function
- Showing improved microglial activity in some brain regions
In other words, the AI prediction was confirmed in a real-world test, at least in this group of mice. The animals did not just “look younger” on molecular measures. They behaved and functioned more like younger mice.
For the field of longevity, this is a big deal. It shows that an AI system can start from messy public data, generate a ranked list of candidates, and actually pick a winner that holds up in a lab experiment.
Why this is exciting, and why you should not rush to take anything
It is very easy to read this and think: “Should I try ouabain?”
The clear answer right now is no.
Here is why:
- The results come from a preprint, which has not yet passed peer review.
- The experiment was done in mice, not humans. Mouse success does not guarantee human safety or benefit.
- Ouabain can be toxic and affects the heart. Self-experimentation with such drugs is dangerous without strict medical supervision.
- The AI system itself still made mistakes, including confusion about control groups and clock outputs in some cases. It is powerful, but not perfect.
The message is “AI can now help find promising drugs much faster, and sometimes its guesses really work in the lab.”
If this approach is refined, AI systems could help:
- Repurpose existing drugs for aging-related diseases
- Prioritize which compounds deserve expensive animal and human trials
- Reveal hidden links between environment, genes, and aging that humans might miss
What comes next for AI in longevity?
🔭 This ouabain result arrives at a time when AI and aging clocks are rapidly improving. Recent reviews of deep aging clocks show that models built from DNA methylation, blood tests, imaging, microbiome data, and more are getting better at predicting health outcomes and mortality risk.
Large multi-omics studies published in 2025 also suggest there are different “types” of aging, with people showing different mixes of metabolic, immune, and inflammatory changes over time. That means future AI systems may not just ask “Does this drug slow aging?” but rather “For which aging type, in which person, under which conditions?”
The ouabain story is likely just one of many test cases to come. Each time an AI prediction is confirmed or disproved, researchers will learn how to tune these systems, which clocks to trust more, and where the pitfalls are.
If this path works, we may see a future where:
- Public data is treated as a gold mine, not a dead archive
- AI agents constantly scan for overlooked combinations of drugs and genes
- Human scientists focus more on testing and understanding mechanisms, guided by AI-generated maps
For now, this study marks a milestone, not a miracle. But for the field of longevity, it is a powerful sign that AI is moving from hype to real, testable progress.
Sources
- Autonomous AI Agents Discover Aging Interventions from Millions of Molecular Profiles – bioRxiv
https://www.biorxiv.org/content/10.1101/2023.02.28.530532v4 - ClockBase – Aging clock analysis platform
https://www.clockbase.org/
