I’ve travelled to many conferences this year and I’ve come across many applications of AI in aging research. One truth echoes: AI can accelerate discovery, but only validated biomarkers make its insights trustworthy.
From AI Models to AI Agents
As highlighted in the recent review on agentic AI for drug discovery by Srijit Seal, Senior AI Scientist at Merck, the field is shifting from static, task-specific models toward autonomous multi-agent systems that can plan, reason, act, and self-correct.
These AI scientists work collaboratively, one agent mines multi-omic data, another designs molecular hypotheses, while others simulate binding, predict toxicity, or generate lab protocols. Together, they can close the loop between hypothesis generation and experimental execution.
Recent demonstrations show that such systems can:
- Generate and validate tens of candidate molecules per day, integrating multi-modal omics and structural data.
- Achieve 400× faster hypothesis testing than manual workflows.
- Drive robotic lab execution through natural-language instructions.
Mining the Molecular History of Aging
At the Biomarkers of Aging conference, Kejun (Albert) Ying, a researcher at Stanford University, presented his project on AI Agent Mining, a multi-agent ecosystem combining millions of public omics samples, including epigenetics from Gene Expression Omnibus (GEO) and beyond. The agents harmonized heterogeneous datasets, recalculated biological age, and ranked interventions by their effect on molecular aging.
Among the rediscovered hits were rapamycin, NAD boosters, and metformin, and uncovered overlooked ones like BRD9 inhibitors and NMNAT overexpression flagged through AI-driven drug repurposing. Each prediction was validated against GenAge and DrugAge databases, underscoring the potential for AI agents to turn dormant public data into new therapeutic hypotheses.
Why Biomarkers Are the Bottleneck
Agentic AI is only as useful as the biological targets it optimises for. Mahdi Moqri et al. have described a framework designed for validating biomarkers for longevity interventions. It makes a simple demand: aging biomarkers must be measurable and reproducible across labs, responsive to known accelerants and geroprotectors, and fit for a defined use (predictive, response, or in rarer cases, surrogate). Until we have response biomarkers that reliably move with real interventions, an AI’s rapid iteration risks overfitting to noise.
The paper also cautions against treating all “clocks” as equal. Estimating how far someone has aged (AgeDev) isn’t the same as how fast they’re aging now (pace), and both can be confounded by tissue choice, batch effects, or transient stressors. The practical path is clear: standardise sampling and assays, prioritise mechanistic and tissue-relevant biomarkers that generalise across cohorts, prove they shift with validated interventions—then let agents optimise against them.
Mitra Bio: Building the Ground Truth for Skin Aging
At Mitra Bio, this is our north star.
We’re building one of the world’s most comprehensive skin methylome databases, using non-invasive sampling and deep sequencing to map how biological age changes under different interventions and over time.
By refining epigenetic clocks and rejuvenation biomarkers specific to skin, we’re creating the kind of reliable, tissue-level biomarkers for meaningful discoveries, not just generating correlations.
In the near future, we envision AI agents mining our skin datasets to uncover new rejuvenating molecules, test topical and systemic interventions, and accelerate translational research in dermatology and longevity.
Mitra Bio has developed a non-invasive skin diagnostics platform that measures biological skin aging. Since skin aging is a highly visible biomarker, tackling it could significantly advance the field of longevity.
If you’re developing longevity ingredients and want to track efficacy using DNA methylation age clocks, let’s connect.
References
- Seal S, Huynh DL, Chelbi M, et al. AI Agents in Drug Discovery. Arxiv. Available online: https://arxiv.org/pdf/2510.27130
- Moqri M, Herzog C, Poganik JR, et al. Biomarkers of aging for the identification and evaluation of longevity interventions. Cell. 2023;Volume 186, Issue 18: 3758-3775
