What Epigenetic Clocks Estimate And How They Are Built

Epigenetic clocks report a biological age that may differ from a person's calendar age. Understanding what that number means requires understanding how the models producing it were built.

The underlying measurement is chemical

DNA carries small chemical additions called methyl groups at specific sites, and these marks influence whether nearby genes are switched on. They are part of how cells maintain their identity.

Methylation at many sites changes in predictable directions across the lifespan. Some positions gain marks steadily with age while others lose them.

The regularity of that drift is what makes prediction possible. It is a measurable signal that correlates with elapsed time without being a designed timekeeping system.

The clock is a fitted statistical model

Building a clock means measuring methylation across many sites in people of known age and fitting a model that selects the most informative sites and weights them.

The output is a weighted sum: a set of sites and coefficients that together predict age. The model is descriptive of its training data rather than mechanistic.

Different research groups using different populations and site selections produce different clocks. Their estimates for the same individual do not always agree closely.

What the model is trained on defines its meaning

First-generation clocks were trained to predict calendar age as accurately as possible, so their output approximates how old a person is.

A clock that predicted calendar age perfectly would carry no additional information. The residual, the difference between predicted and actual age, is where the interest lies.

Later clocks were trained instead against health outcomes or mortality risk, meaning their output is a risk-related score expressed in years rather than an age estimate.

Tissue and technique affect the reading

Methylation patterns differ between tissues, so a clock developed on blood may perform poorly on another tissue. Some clocks are explicitly designed to work across tissues.

Blood samples contain a mixture of cell types whose proportions shift with age and illness. A change in that mixture can move the clock reading without any change within individual cells.

Measurement noise is also significant relative to the differences being detected. Repeat samples from the same person on the same day can differ by a meaningful margin.

Correlation is not yet causation

A clock reading above calendar age is associated with elevated risk at population level, which is what training against outcomes was designed to achieve.

Whether methylation drives aging or records it remains unresolved. The models were built for prediction, and prediction does not establish direction.

Commercial tests reporting a biological age are therefore reporting a model output, not a clinical measurement. Health concerns are properly assessed by a clinician using established diagnostics.