What are biological age and biomarkers of aging?

Chronological age is the time since birth. Biological age is a broader concept: an attempt to describe age-related changes in the body and the associated differences in health. There is not yet a single widely accepted measure that captures every organ and dimension of aging. People of the same age can differ considerably in health, and different tissues within one person can change differently. [1] [5]

A biomarker is a measurable sign of a biological process. To study aging, researchers use blood measures, proteins and features of DNA, among other markers. A model combines selected traits into an estimate, but that estimate describes only what its data and method can help capture.

How epigenetic clocks work—and where AI fits in

Among the most studied biomarkers are changes in DNA methylation: chemical marks at specific places on the molecule. These marks can help regulate how genes work without changing the DNA sequence. Epigenetic clocks use patterns in such marks to calculate a score from a sample, often blood. For example, Horvath’s clock was originally trained to predict chronological age using data from different tissues. [1] [2]

Clock development uses statistics and machine learning, one approach within AI. An algorithm fits relationships between the data and its training target, then calculates an estimate for a new sample. Not all clocks use neural networks. However sophisticated the algorithm, its output depends on measurement quality, the training data and precisely what it was trained to predict. [5]

What a result means—and what it does not show

If a model gives an age in years, a number above or below chronological age means that the model’s estimate differs from the person’s actual age. It does not literally mean that every organ has become older or younger by that number of years. A blood test reflects the features of a particular sample and is not a complete assessment of the body. [2] [5]

In research, some clocks are associated with the likelihood of age-related diseases or mortality across groups of people. For example, DNAm PhenoAge has been associated with all-cause mortality. Such population-level associations do not allow a single result to establish a diagnosis, name a date of death or guarantee any one person’s future health. [3]

Why different clocks do not agree

Clocks answer different questions. Horvath’s model was created to estimate chronological age from DNA methylation. DNAm PhenoAge was trained against a measure combining age and clinical indicators linked to mortality. DunedinPACE was developed to estimate the pace of age-related change using long-term observations of physiological measures; its result is a pace score, not an age in years. These outputs cannot simply be compared as if they were the same measurement. [2] [3] [4]

Results are also affected by the DNA sites selected, tissue type, cell composition, laboratory methods and the group whose data were used to train the model. A difference between tests does not necessarily mean one is wrong: they may measure different aspects. Even a repeat test may differ because of measurement variation rather than a reliable change in health. [4] [5]

Correlation, model validation and clinical benefit

Correlation means that measures vary together; on its own, it does not prove that one causes the other. An association between a clock and mortality in an observational study does not prove that changing the clock’s reading will extend life. Other factors may influence both outcomes. [5]

A reliable test must measure its marker reproducibly. Researchers then need to check how well the model performs on independent data and in different groups, and whether it adds useful information to measures already available. Clinical benefit requires a separate test: do real outcomes improve when the test is used in medical decisions? Accurate age prediction alone does not establish that. [5]

Can a personal result be improved, and can life be extended?

A change in a clock score is a potential surrogate marker, not a proven substitute for a clinical outcome such as disease incidence or lifespan. In the randomized CALERIE trial of adults without obesity, a subsequent analysis found a small reduction in the DunedinPACE score in the calorie-restriction group compared with the control group. There were no significant changes in the PhenoAge and GrimAge clocks. The study did not prove that human lives were extended. [6]

Experiments in cells and animals help researchers study mechanisms, but their results cannot automatically be applied to humans. It remains unclear which changes in a particular clock reliably indicate better health for an individual, how lasting they are and whether they lead to a longer life. The CALERIE findings are not a personal recommendation to restrict food intake. [1] [5] [6]

Questions to ask before taking a test

It is useful to find out the model’s name, the sample type and exactly what it calculates. Ask whether the method has been tested in independent groups, how reproducible the result is and whether its uncertainty is known. For repeat measurements, comparable methods and laboratory conditions matter. [5]

A separate question is whether the result will support a better decision than the health information already available. Be especially cautious if a test comes with promises of guaranteed rejuvenation or products offered on the basis of one number. Discuss the medical meaning of results with a qualified professional rather than treating them as a diagnosis on their own.

The takeaway

Epigenetic clocks and AI help researchers explore the complex patterns of aging, but they cannot capture an entire health history in one number. For now, their most established role is in research; moving from a change in a biomarker to a proven improvement in people’s lives requires further clinical testing.

Sources

  1. National Institute on Aging, National Institutes of Health — The epigenetics of aging: What the body’s hands of time tell us (2021)
  2. Horvath S. — DNA methylation age of human tissues and cell types, Genome Biology (2013)
  3. Levine ME et al. — An epigenetic biomarker of aging for lifespan and healthspan, Aging (2018)
  4. Belsky DW et al. — DunedinPACE, a DNA methylation biomarker of the pace of aging, eLife (2022)
  5. Moqri M et al.; Biomarkers of Aging Consortium — Validation of biomarkers of aging, Nature Medicine (2024)
  6. Waziry R et al. — Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial, Nature Aging (2023)