Skip to content

Epigenetic clock test: what it measures and how to run one on your own data

Oak
A huge fog-shrouded animal whose hide is studded with thousands of pale glowing beads along dark filaments, with smaller sparsely beaded animals nearby.

An epigenetic clock test measures DNA methylation at a few hundred to a few thousand CpG sites and feeds those values into a linear model trained to predict chronological age or a health outcome. The output is a number in years, or in the case of pace-of-aging clocks a ratio near 1.0. The measurement is real and reproducible within known error bars. The interpretation is much weaker than the marketing suggests: a single clock reading tells you very little, and the difference between two readings taken months apart is often dominated by assay noise and blood cell composition rather than by anything about you.

We think the test is worth running if you treat it as one longitudinal biomarker among many and you keep the raw data. It is not worth running if you want a single number to act on.

What the clock is measuring

DNA methylation is the addition of a methyl group to cytosine, almost always at CpG dinucleotides in mammalian genomes. At a given site in a given sample, you measure a beta value between 0 and 1: the fraction of DNA molecules in that sample that carry the methyl mark. Beta drifts with age at thousands of sites, in both directions, and the drift is regular enough that a penalized regression on a few hundred sites predicts chronological age in blood to within about three to four years.

That regularity is the entire basis of the field. First-generation clocks such as Horvath’s multi-tissue clock (353 CpGs) and Hannum’s blood clock (71 CpGs) were trained directly on chronological age. Second-generation clocks were trained on something else. PhenoAge (513 CpGs) was trained on a composite of clinical chemistry values and mortality. GrimAge was trained on methylation surrogates for plasma proteins and smoking pack-years. DunedinPACE (173 CpGs) was trained on the rate of change in a panel of organ-function measures over two decades, so it reports a pace rather than an age.

The distinction matters for interpretation. A first-generation clock that says you are older than your birth certificate is telling you that your methylation pattern looks unusual for your age. It is not measuring damage, and the mechanistic link between the clock sites and any aging process remains unresolved 1. Second-generation clocks are better predictors of morbidity and mortality because they were trained on those outcomes, which also means they partly encode things you already know, such as whether you smoke.

The assay sets your noise floor

There are three ways to get the methylation values, and the choice determines what you can compute and how much the number will wobble.

The default is an Illumina methylation BeadChip. The current version is EPIC v2, roughly 935,000 probes. Its predecessors were EPIC v1 (about 850,000) and 450k (about 485,000). Arrays are cheap, the clock coefficient sets were derived on them, and every published clock has an array implementation. The catch is that EPIC v2 dropped and redesigned probes relative to v1, so some legacy clock CpGs are absent and must be imputed from training-set means. Every imputed site adds bias you cannot see in the output.

The second route is bisulfite sequencing. Whole-genome bisulfite sequencing gives you every CpG at whatever depth you pay for, and clocks can be trained natively on that resolution rather than retrofitted from array coefficients 2. It is far more expensive per sample and the coverage per site is low unless you sequence deeply, which is a problem because a beta value from 8 reads is a coarse estimate.

The third route is targeted sequencing of the clock CpGs only. Multiplex amplicon assays can quantify the specific cytosines behind several published clocks in one reaction, at a fraction of array cost 3. Single-locus tests take this further: an ELOVL2 promoter assay read out by next-generation sequencing predicts age from one region in saliva or blood 4. Newer library methods reduce the cost and hands-on time of methylation measurement enough to make repeated sampling practical 5. The tradeoff is that a targeted panel locks you into the clocks it covers, and you cannot recompute a clock published next year.

One thing worth being clear about: standard short-read whole-genome sequencing gives you no methylation information at all, because bisulfite conversion or enzymatic conversion is what encodes the mark into the sequence. Nanopore sequencing is the exception. If you have Oxford Nanopore reads basecalled with a modification-aware model, you already have 5mC calls, and modkit pileup will give you per-CpG methylation fractions directly from the BAM.

Running it yourself

If you have array IDATs, the whole thing runs in R in about twenty minutes. We would use sesame for preprocessing over minfi, mainly because its masking of problematic probes is more current for EPIC v2.

library(sesame)
betas <- openSesame("idats/", prep = "QCDPB", func = getBetas)

QCDPB applies quality masking and detection p-value masking. It then applies out-of-band background correction (noob), dye bias correction, and pOOBAH. Check the failure rate before you go further. Drop any sample where more than about 1 percent of probes fail detection at p < 0.01, and inspect the intensity distributions rather than trusting the pipeline silently.

Then compute the clocks. The methylclock package covers Horvath, Hannum, PhenoAge, skin-and-blood, and a few others from a matrix of betas with CpG identifiers as row names. DunedinPACE has its own package and its own gold-standard normalization step, which you should not skip. Also estimate cell composition with FlowSorted.Blood.EPIC and the IDOL probe set, and record the estimates alongside the clock outputs.

Finally, prefer the principal-component versions of the clocks where they exist (PCHorvath, PCPhenoAge, and PCGrimAge). They project the betas onto principal components before applying the regression, which averages away much of the probe-level technical noise. Test-retest reliability improves substantially, and reliability is the binding constraint if you plan to track yourself over time.

What moves the number that has nothing to do with you

Cell composition is the largest confounder in blood. Granulocytes, CD8 T cells and B cells have distinct methylomes, so a shift in your differential from a recent infection moves the clock. Always report cell fractions with the clock value, and look at the residual after adjusting for them.

Clonal hematopoiesis is the second. Blood is produced by a finite pool of stem cell clones, and that clonal structure changes sharply in later life as a few clones come to dominate 6. When the clone mix changes, the methylation profile of the tissue you sampled changes with it. More broadly, somatic mutation and mosaicism accumulate across tissues with age and alter what a bulk sample represents 7.

Then there is the mundane layer: batch, chip position, DNA input, time between draw and extraction. If you want a longitudinal series, bank aliquots and run them on the same platform, ideally in the same batch. Two samples run a year apart on different platforms are barely comparable.

Cost, and whether to bother

Consumer epigenetic age tests run roughly $200 to $500 for a saliva or blood spot and a PDF report. Research-grade EPIC array processing through a core facility is often $300 to $600 per sample, and you get the IDATs. We would always pay for the IDATs. The report is a linear model you can recompute yourself, and the coefficients are published.

Our view: run it at baseline, run it again at twelve months on the same platform, and treat any change smaller than the reported test-retest error as no change. Do not use the result to make health decisions. If a clock reading concerns you, the useful next step is standard clinical workup with a physician. Another clock is not the answer.

Questions people also ask

At what age do you age most rapidly? Methylation age advances fastest in early childhood, which is why Horvath’s model uses a log transform for ages under 20. In adulthood the relationship is close to linear. Reports of nonlinear waves of molecular change in midlife exist, but they come from other omics layers and are not well established in the clock literature.

How much does an epigenetic test cost? Consumer tests are roughly $200 to $500. Array processing with raw data delivery is roughly $300 to $600. Targeted amplicon and pooled library methods are pushing the per-sample cost well below that in research settings 5.

Are epigenetic clocks accurate? They predict chronological age in blood to within about three to four years, which is accurate as regression goes. Accuracy against biological aging is a different question, and the causal interpretation of the clocks is unsettled 1. Reproducibility on the same sample is the practical limit, and it is why we recommend principal-component clock variants.

Are epigenetic tests worth it? Worth it as a repeated measurement with the raw data in hand. Not worth it as a one-off number, and not worth it as a basis for any decision you would otherwise take to a clinician.

Footnotes

  1. Calen P. Ryan. “Epigenetic clocks”: Theory and applications in human biology. American Journal of Human Biology, 2020. https://doi.org/10.1002/ajhb.23488 ↩ ↩2

  2. Congcong Hu, Yunxiao Li, Longhui Li, et al. BS-clock, advancing epigenetic age prediction with high-resolution DNA methylation bisulfite sequencing data. Bioinformatics, 2024. https://doi.org/10.1093/bioinformatics/btae656 ↩

  3. Ewelina Pośpiech, Aleksandra Pisarek, Joanna Rudnicka, et al. Introduction of a multiplex amplicon sequencing assay to quantify DNA methylation in target cytosine markers underlying four selected epigenetic clocks. Clinical Epigenetics, 2023. https://doi.org/10.1186/s13148-023-01545-2 ↩

  4. David Cheishvili, Sonia Do Carmo, Filippo Caraci, et al. EpiAge: a next-generation sequencing-based ELOVL2 epigenetic clock for biological age assessment in saliva and blood across health and disease. Aging, 2025. https://doi.org/10.18632/aging.206188 ↩

  5. Patrick T. Griffin, Alice E. Kane, Alexandre Trapp, et al. TIME-seq reduces time and cost of DNA methylation measurement for epigenetic clock construction. Nature Aging, 2024. https://doi.org/10.1038/s43587-023-00555-2 ↩ ↩2

  6. Emily Mitchell, Michael Spencer Chapman, Nicholas Williams, et al. Clonal dynamics of haematopoiesis across the human lifespan. 2021. https://doi.org/10.1101/2021.08.16.456475 ↩

  7. Elizabeth Pan, Alexander Maslov, Jan Vijg. Somatic mutations and genome mosaicism in aging and disease. Experimental & Molecular Medicine, 2026. https://doi.org/10.1038/s12276-026-01791-3 ↩

Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.