In short
An epigenetic clock is a mathematical model that estimates biological age from patterns of DNA methylation — chemical tags added to DNA that change predictably with age — at a defined set of genomic sites.
DNA methylation is an epigenetic modification (it changes gene activity without altering the DNA sequence itself) that adds a methyl group to cytosine bases, typically at CpG sites. Methylation at certain CpG sites changes in a remarkably consistent, quantifiable way as a person ages, and an epigenetic clock is a statistical model — usually built with penalized regression on methylation array data — trained to predict age (or an age-related outcome) from methylation levels at a selected panel of these sites.
The field's foundational model is the Horvath clock, published by Steve Horvath in 2013, trained on chronological age across 51 tissue types. Later clocks changed what they optimize for: PhenoAge (Levine et al., 2018) was trained to predict a composite of clinical biomarkers and mortality risk rather than chronological age alone, and GrimAge (Lu et al., 2019) was trained directly on time-to-death and time-to-disease-onset data, incorporating estimated levels of plasma proteins and smoking pack-years. In published cohort studies, GrimAge and PhenoAge have shown stronger associations with mortality and disease incidence than the original Horvath clock or than chronological age alone.
What these clocks have NOT been shown to do is serve as a validated, FDA-cleared surrogate endpoint for anti-aging drug trials, and results vary meaningfully between commercial testing platforms because each uses its own proprietary algorithm and reference population. A given supplement or lifestyle change lowering someone's "epigenetic age" on one commercial test is a biomarker change, not proof of a change in actual disease risk or lifespan — that link is inferred from population-level mortality correlations, not demonstrated for each individual intervention.
Worth remembering
- Estimates biological age from DNA methylation patterns at specific genomic sites.
- Horvath (2013), PhenoAge (2018), and GrimAge (2019) are the major published clocks, each trained on a different target.
- Commercial test results are not interchangeable and are not a validated proxy for an individual's disease risk.
Where this shows up on the site
Frequently asked questions
Which epigenetic clock is most accurate?
It depends what "accurate" means for the question being asked. GrimAge and PhenoAge correlate more strongly with mortality and disease incidence in published cohorts than the original Horvath clock, which was trained on chronological age rather than health outcomes.