Polypharmacology meets machine learning as researchers harness AI to discover multi-target drugs that slow aging in worms.
For years, the prevailing ethos in pharmacology has been one of precision – identify a single target, design a molecule to engage it and fine-tune that interaction until it behaves predictably. This model has yielded many successes, not least in oncology and infectious diseases, but for the diffuse, entangled biology of aging, its effectiveness may have run its course. Now, a new study, published in Aging Cell, suggests that targeting the multiplicity of aging processes might be more fruitful – and that artificial intelligence is ready to lend a hand.
Aging may be systemic – and drug discovery could be catching up
The research, a collaboration between Singapore-based Gero and Scripps Research in California, combines computational modeling and in vivo testing in Caenorhabditis elegans to identify and validate compounds that act on multiple biological pathways [1]. Central to the study is polypharmacology – the deliberate design of compounds that interact with several molecular targets at once. Once considered too complex and risky to pursue, this approach may now be viable, thanks to modern AI methods capable of discerning patterns in sprawling pharmacological data.
“It’s not just an incremental step. This is a genuine step change,” said Dr Michael Petrascheck of Scripps, who co-led the study with Gero’s Peter Fedichev. “It shows that AI can help researchers tackle exponentially more complex biological questions than they could have unassisted.”
Graph neural networks and GPCRs
The team employed graph neural networks (GNNs), using a model called Chemprop, to predict compounds likely to bind across specific clusters of G-protein coupled receptors (GPCRs) [1]. These receptors are often treated warily in drug development due to concerns over off-target effects, yet they represent a crucial interface between extracellular signals and intracellular responses – and, as it turns out, a productive point of intervention for aging.
Using binding data from the ChEMBL database and the ZINC20 compound library, the AI predicted hundreds of candidates. Twenty-two novel compounds were then selected and tested in C elegans; of these, 16 significantly extended lifespan. Twelve produced increases of over 40%, while one compound – ZINC000019802386 – extended lifespan by a whopping 74%, placing it among the top-performing geroprotective agents in the DrugAge database [1].
The analysis suggested that the most effective compounds bound simultaneously to three GPCR clusters: drd2 (dopaminergic), hrh1 (histaminergic) and htr6 (serotonergic). These findings were supported by follow-up experiments using receptor knockouts, confirming that the longevity effects were dependent on the presence of multiple GPCR targets.
A strategic embrace of complexity
“Traditional drug discovery obsesses over precision, aiming to modulate a single pathway with laser-like focus,” said Fedichev. “But aging doesn’t work that way. It’s systemic, intertwined, and defies one-dimensional solutions. That’s what our approach embraces.”
While multi-target drugs have shown efficacy before, this study represents the first known instance of polypharmacological compounds for aging being intentionally designed from the outset using AI – and successfully validated in vivo. Historically, drugs exhibiting polypharmacological effects have emerged by accident – their secondary effects tolerated or, occasionally, fortuitous. Here, such effects are the goal rather than the risk.
Longevity.Technology: This valuable study marks a decisive shift in longevity drug discovery – from the pursuit of pinpoint precision to an embrace of biological complexity. By leveraging AI to design compounds that act on multiple GPCR targets simultaneously, the research team has not only delivered a remarkable 70% hit rate for lifespan extension, but also achieved some of the largest effect sizes seen to date in C elegans. Such outcomes are not just incremental improvements – they signal that polypharmacology, when intentionally and computationally designed, may be key to unlocking the systemic nature of aging. In moving away from the old paradigm of single-target “magic bullets,” this work opens a pathway to therapies that reflect the multifactorial decline seen in biological aging.
What makes this approach even more compelling is its scalability. By integrating deep pharmacological datasets, graph neural networks, and large compound libraries, the researchers have built a generalizable pipeline – one capable of exploring chemical space far more efficiently than trial-and-error methods. This model doesn’t just predict hits; it predicts networks of action, and does so with a level of confidence that can now be validated in vivo. It’s a step toward not just accelerating drug discovery, but transforming it. If aging is a systemic breakdown of resilience, then AI-driven polypharmacology may be the systems medicine we’ve been waiting for – we sat down with Peter Fedichev to find out more.
Intentional polypharmacology – and a path beyond worms
“Traditional drug discovery obsesses over precision, aiming to modulate a single pathway with laser-like focus,” said Fedichev. “But aging doesn’t work that way. It’s systemic, intertwined, and defies one-dimensional solutions. That’s what our approach embraces.”
Still, while the worm data are striking, Fedichev is clear-eyed about what lies ahead. “The work serves as a technology demonstrator,” he said. “We’ve shown for the first time that compounds with polypharmacological activity can have significant effects on longevity, at least in simpler organisms like C elegans.”
The compounds identified in the study include several neuroleptics – drugs more familiar to psychiatrists than geroscientists. “Obviously that context doesn’t directly apply to roundworms,” Fedichev noted, “but it highlights how unexplored the pharmacological landscape of aging still is.” Importantly, these compounds have not been tested in vivo in mammals – a key translational gap. “Mammals have hearts, unlike roundworms,” he said. “So if the life-extension effects are confirmed in higher organisms, these compounds would need significant optimization to address liabilities like cardiotoxicity.”
From mechanism to medicine
Asked whether their AI model considered pharmacokinetic factors like absorption and toxicity, Fedichev clarified that while ADMET properties weren’t built directly into the machine learning algorithm, downstream filtering was applied. “We identified available molecules that could exert polypharmacological effects on longevity in these animals,” he said. “Then we filtered based on known drug-like properties – including toxicity and bioavailability – to ensure that the proposed compounds could effectively penetrate into the roundworms and function in vivo.”
For future iterations, integrating these constraints more directly into the AI model may prove important – particularly for advancing beyond early-stage discovery.
The case for complexity
The study’s polypharmacological rationale raises a larger question: is the era of single-target longevity drugs drawing to a close? “In a sense, any biological system is inherently interconnected,” Fedichev said. “Even targeting a single molecule often leads to downstream effects on many transcripts, proteins and metabolites.”
Yet he argues that their findings suggest more than just interconnection. “Most biological systems have evolved a kind of robustness,” he said. “That complexity limits the effectiveness of one-target-one-disease strategies. Aging, like many chronic diseases, doesn’t have a single root cause – so it’s not easily undone by simple interventions.”
That said, he isn’t entirely dismissive of more traditional approaches. “Single-target drugs have practical advantages – they’re easier to develop, understand and regulate. But if we want to make real progress in treating aging, we’ll need to embrace systems-level, polypharmacological approaches. I believe that’s not just a promising direction – it’s the future of medicine.”
Tuning the network
Finally, the study suggests that not just which receptors are targeted matters, but how. The variable effects of lifespan extension observed with ziprasidone and zotepine in receptor knockouts hint at ligand bias – the idea that different compounds binding to the same receptor can elicit different downstream responses.
Fedichev sees this as a fertile area for future refinement. “Many successful drugs are inherently polypharmacological,” he said. “It’s one of the reasons small molecules continue to thrive, even in the era of biologics and gene therapies.”
He points to recent advances in AI – such as AlphaFold and improved molecular modelling tools – as enablers of the next generation. “We’re entering an era where polypharmacology can be not only described but predicted with high precision,” he said. “That opens the door to small-molecule medicines designed to hit multiple targets within key pathways – acting on networks, not just nodes.”
In doing so, such compounds may not only prove more effective, but also more resilient. “Even in the face of mutations or variable disease mechanisms,” he added, “they can generate more robust responses. That’s especially important for aging – a condition that resists simple answers.”
While the study is limited to C elegans, and many questions remain about translational relevance and toxicity in more complex organisms, the methods employed point to a new framework for discovery. One that is not afraid of complexity – and may, at last, be able to make sense of it.
[1] https://onlinelibrary.wiley.com/doi/10.1111/acel.70060
The post AI-designed compounds show promise in lifespan extension appeared first on Longevity.Technology – Latest News, Opinions, Analysis and Research.
