How AI Is Finding New Molecular Targets for Healthy Aging

Scientists studying aging have a new tool: artificial intelligence systems that can scan thousands of molecules or genetic datasets in days and flag the handful most likely to affect how cells age. Several projects made public in recent weeks show how this works in practice — among them a Brazilian university lab, an international biotechnology company, and a research team mining large public genomic databases — and they all aim at the same underlying problem: oxidative stress and cellular damage that accumulate with age.

An explainable AI model hunts for antioxidant peptides

At the Pontifícia Universidade Católica do Rio de Janeiro (PUC-Rio), a team led by chemistry professor André Silva Pimentel, working with researchers Maria Carolina J. A. Schneider and Marina Geisiely Damaso Godinho Alves, published a study on explainable artificial intelligence applied to antioxidant peptides, according to an article the team wrote for The Conversation Brasil on October 5, 2026. Peptides are short chains of amino acids that occur naturally in food proteins or can be synthesized, and some of them can neutralize free radicals — unstable molecules that damage cells when they build up faster than the body's antioxidant systems can clear them.

The PUC-Rio group used a method called explainable AI, which differs from standard machine-learning models in one key respect: instead of only producing a prediction, it also shows which parts of a peptide's amino-acid sequence most influenced that prediction. According to the researchers, this lets scientists see which structural features make a peptide more likely to act as an antioxidant, rather than treating the AI as a black box. The work was funded by FAPERJ, Rio de Janeiro's state research foundation, and was also covered by the Brazilian outlet Brasil em Folhas, which described the same laboratory and methodology.

The practical value, the researchers wrote, is speed: instead of testing large numbers of peptide candidates one by one in a lab, the AI model narrows a long list down to the few sequences most worth testing experimentally. The team said potential applications extend beyond anti-aging cosmetics and nutraceuticals to research on oxidative-stress-linked conditions such as Alzheimer's disease, Parkinson's disease and cancer — though they described this as a direction for further study, not a finished treatment.

Larger AI systems are scanning genomes and drug libraries for aging targets

A separate, larger-scale effort comes from Insilico Medicine, a biotechnology company that develops AI tools for drug discovery. On September 17, 2026, the company published a study on the cover of the journal Cell describing an open-source toolkit for longevity research, built with Liquid AI, the Buck Institute for Research on Aging, Harvard Medical School and Brigham and Women's Hospital, according to Insilico Medicine's own announcement.

The toolkit has three parts: LongevityBench, a benchmark that tests how well AI models reason about aging biology across five data types including genetics and proteomics; Longevity-LLMs, a family of compact language models (ranging from 0.6 to 9 billion parameters) trained specifically on aging-related data; and Longevity Claw, an autonomous research platform meant to identify and rank possible aging-related targets. Insilico Medicine reported that its specialized 9-billion-parameter model matched or exceeded 16 frontier general-purpose AI systems on aging-biology tasks, and that Longevity Claw proposed 328 candidate aging-intervention targets, which it said showed enrichment of up to 5.6-fold for aging-related genes when compared with an independently published reference set of experimentally validated aging targets.

A third approach, described by the aging-research news site Lifespan.io, used an AI tool called ClockBase Agent — built by a team led by researcher Vadim Gladyshev — to mine over 2 million existing RNA-sequencing and epigenetic datasets covering 43,529 recorded biological interventions. The system flagged known drugs including rapamycin, metformin, fenofibrate and ouabain as linked to slower biological aging in the data; Lifespan.io reported that in follow-up tests on 20-month-old mice, three months of ouabain treatment was associated with improved frailty scores, cognitive performance, fur condition and heart function, though the preprint describing the full study was posted in November 2025 and has not been confirmed in human trials.

What this research does and does not show

None of these projects claims to have produced an approved anti-aging treatment. The PUC-Rio peptide model and Insilico Medicine's target list are prediction tools meant to prioritize candidates for laboratory and clinical testing, not finished therapies. The ouabain findings described by Lifespan.io come from animal experiments, not human trials. Researchers involved in the PUC-Rio study and Insilico Medicine both frame their work as narrowing a very large search space — millions of possible molecules or gene targets — so that scarce lab and clinical resources are spent testing the most promising candidates first.

• Explainable AI (PUC-Rio): flags which amino-acid sequences in food-derived or synthetic peptides are likely antioxidants.
• LongevityBench / Longevity-LLMs / Longevity Claw (Insilico Medicine, with Buck Institute and Harvard Medical School): benchmark and rank genetic targets linked to aging across omics datasets.
• ClockBase Agent (Gladyshev research team): mines existing public genomic datasets to flag existing drugs associated with slower biological aging in animal data.