This is a step change for drugs against aging

🤙🏼 Want to connect? Add me on LinkedIn. 🙏🏼 Not subscribed to the LEVITY podcast on Youtube yet? Do it here. 🎧 More of a listener? The LEVITY podcast is also available on Spotify, Apple Podcasts and other places.

🏡 Quick housekeeping: the regular LEVITY newsletter will land in your inbox tomorrow, complete with all the usual sections. But the team at Gero reached out to give me a heads-up about a fascinating new study - and I decided it deserved its own “special report.” This is genuinely spectacular stuff.

New study shows machine learning can design multi-target drugs to tackle aging’s systemic complexity - and they actually work.

Scientists often call them dirty drugs. Compounds that bind to multiple targets, meddling with entire systems instead of acting with laser precision. In traditional drug development, this kind of behavior is a red flag. It suggests unpredictable effects, unwanted side actions, regulatory headaches. So pharmaceutical companies have mostly discarded them.

But what if dirty drugs are, in fact, potent weapons against aging?

A new study published in Aging Cell by researchers from Gero - including recent LEVITY podcast guest Peter Fedichev (pictured above) - and Scripps Research challenges the core assumptions behind modern drug discovery.

Using an AI-driven approach, the team did something few believed was possible: they intentionally designed polypharmacological drugs - compounds that hit multiple targets simultaneously - with the specific goal of extending lifespan.

Oh, and one more little detail: it actually worked.

And not just marginally. One of the AI-designed molecules extended the lifespan of household aging model, the roundworm C. elegans, by 74%, placing it among the most effective life-extending compounds ever discovered in this model organism. Even more striking: over 70% of the tested candidates worked, a hit rate far beyond traditional screens.

“It’s not just an incremental step. This is a genuine step change,” said Dr. Michael Petrascheck, professor at Scripps Research, and one of the authors behind the study, in a press release.

Dr. Matt Kaeberlein, leading rapamycin expert and co-founder of Ora Biomedical*, a company focusing on accelerating the discovery of longevity therapeutics by combining AI, robotics, and high-throughput screening in C. elegans, said: “This study highlights the enormous potential of AI to accelerate the discovery of new longevity therapeutics by identifying interventions that act on multiple targets simultaneously.”

Kaeberlein, however, also stressed the need for “higher quality, mechanistically annotated datasets across diverse models and interventions”.

* Related to this is my recent interview with Krister Kauppi.

Traditional pharma favors the “magic bullet”: one molecule, one target, one effect. But aging isn’t a single malfunction. It’s a systemic breakdown - a collapse of interdependent processes across networks of pathways.

As Peter Fedichev puts it:

“[Aging] is systemic, intertwined, and defies one-dimensional solutions. That’s what our approach embraces.”

The Gero–Scripps team trained a graph neural network not to avoid multi-target binding, but to seek it out. They built a model that predicts which combinations of targets are most likely to produce longevity effects when hit together. Then they searched vast chemical libraries for compounds that matched. The resulting “dirty drugs,” rather than being messy, were highly effective - designed messiness, not accidental.

Sure, some of the world’s most trusted drugs are dirty. Metformin, for example - the widely used diabetes drug and a by some seen as a candidate for slowing aging - hits multiple targets. In fact, scientists doesn’t really know how the heck it works. And yet it’s considered safe, effective, and even boring by some.

In other words: the issue with dirty drugs isn’t promiscuity. It’s the lack of intentionality.

What this study demonstrates is that AI can now do what old-school medicinal chemistry couldn’t: design these multi-target drugs on purpose, with clear therapeutic direction.

This, by the way, inverts the usual critique of AI in drug discovery. Many in the longevity field are dismissive of AI, seeing it as hype and hopium. But here, AI solved the part of biology that biology - our human brains - couldn’t. Polypharmacology has long been considered too complex to design for - too many variables, too much noise. Now, with the right data and architecture, the “impossible” becomes tractable.

“What’s exciting is that this approach is the idea that to target aging, we might have to change our approach to drug discovery by rationally impacting multiple targets at once. It is an elegant and potentially scalable strategy that could help shape future therapies to improve human healthspan as well”, said Jan Gruber, Associate Professor at Yale-NUS.

While C. elegans is partly used for convenience - it’s short-lived, low-cost, and easy to test in high numbers - it also happens to be one of the only organisms with enough detailed lifespan and pharmacology data to train AI on complex, multi-target interventions. The “scalable strategy” Gruber refers to isn’t about applying worm results to humans of course, but more likely about building a repeatable design logic: using AI to identify combinations of targets worth hitting, and finding compounds that hit them together.

If the researchers are right - and the data suggests they might be - then the future of geroprotection may depend less on what we can isolate and more on what we can orchestrate. And maybe the drugs we once called dirty are exactly what we need to clean up aging.