What an Epigenetics Testing Kit Measures, and What the Numbers Mean

Oak
epigenetics methylation biological-age bioinformatics assays
A giant felled glowing tree trunk in a bioluminescent forest, its rings studded with tiny bright and dark mineral beads.

An epigenetics testing kit is a sample collection tube plus an array assay. You spit in a tube or swab your cheek, a lab does bisulfite conversion on the extracted DNA and hybridizes it to an Illumina Infinium array (EPIC v1, ~866,000 CpG sites, or EPIC v2, ~936,000), and a software pipeline turns per-CpG methylation fractions into a handful of numbers: an epigenetic age, a pace-of-aging rate, sometimes deconvolved immune cell proportions. Consumer kits run about $200–$500. Research-grade array processing at a core facility is $150–$300 per sample, plus DNA extraction. The measurement is real and reproducible at the CpG level. The interpretation layer sitting on top of it is where the uncertainty lives, and most vendors do not show you the error bars.

What is physically happening in the assay

Methylation lives on the 5-carbon of cytosine, almost always in a CpG dinucleotide. The array cannot read a methyl group directly, so the lab converts it to a sequence difference. Sodium bisulfite deaminates unmethylated cytosine to uracil (read as T after PCR) and leaves 5-methylcytosine alone. Every CpG then has two possible sequences, and the array probes measure the ratio.

You get two intensities per probe, methylated (M) and unmethylated (U). Beta value is M / (M + U + 100), bounded 0 to 1, interpretable as the fraction of cells in your sample where that site is methylated. M-values are log2(M/U), which are better behaved for linear models because betas are heteroscedastic near 0 and 1.

Two probe chemistries coexist on the Infinium arrays. Type I probes use two beads per CpG in the same color channel; Type II use one bead and two colors. Their beta distributions differ, which is why every pipeline includes a normalization step (BMIQ, noob, or funnorm). This choice is not cosmetic. McEwen and colleagues evaluated preprocessing pipelines against epigenetic age estimation on the EPIC array and found that the choice of normalization meaningfully shifts the resulting age estimates, with no single method dominating across clocks.1 If a vendor changes pipelines between your first and second test, your “aging” may be entirely pipeline drift.

The alternative platform is targeted: instead of a million sites, you amplify and quantify a dozen. A recent dodecaplex droplet digital PCR assay quantifies methylation at 12 clock CpGs with sub-percent precision and produces age predictions without array normalization at all.2 For longitudinal self-tracking, that tradeoff is attractive. Fewer sites, but far less pipeline-dependent variance between runs.

Sample type changes the answer

Blood and saliva are not interchangeable. Whole blood methylation is dominated by leukocyte composition: a shift in neutrophil-to-lymphocyte ratio moves thousands of CpGs, and clocks trained on blood partly read cell composition. Saliva is a mixture of buccal epithelium and leukocytes in a ratio that varies with how you collected the sample, the time since you ate, and whether you had a cold. Eipel and colleagues showed that epigenetic age predictions from buccal swabs become substantially more precise once you adjust for cell-type-specific methylation signatures, because the epithelial-to-leukocyte fraction is the dominant confound.3

Practical consequence: if you are going to track methylation over time, fix the sample type, the collection time of day, and the lab. Compare blood to blood. Run a CBC with differential on the same draw so you can regress out cell composition yourself, or use a reference-based deconvolution (Houseman method, FlowSorted.Blood.EPIC in Bioconductor) and check that the estimated proportions match the CBC.

Working with the raw data

Insist on IDAT files. Two per sample, _Grn.idat and _Red.idat, roughly 8 MB each. Without them you are holding a PDF. Vendors that will not release IDATs are selling you a score, not a measurement.

A minimal pipeline in R:

library(minfi); library(sesame)
rgset <- read.metharray.exp(base = "idats/")
# QC first: detection p-values per sample
det <- detectionP(rgset)
colMeans(det > 0.01)          # drop samples > 1% failed probes
# Normalization; noob for single samples, funnorm for batches
mset <- preprocessNoob(rgset)
beta <- getBeta(mset)
# Drop cross-reactive and SNP-overlapping probes (Pidsley/Zhou lists)

Sanity checks before you trust anything downstream. Bisulfite conversion controls should sit above 80% converted. Predicted sex from X/Y probe intensity should match yours. The 59 SNP probes on the EPIC array give you a genotype fingerprint, so a follow-up sample from the same person should correlate above 0.99 with the first one. If it does not, a sample got swapped.

Clock implementations are public. methylclock and dnaMethyAge in R cover Horvath 353-CpG multi-tissue, Hannum blood, PhenoAge, GrimAge variants, and DunedinPACE. Run several. Where they disagree by more than a couple of years, that disagreement is the real uncertainty in your estimate.

What the number is telling you

First-generation clocks (Horvath, Hannum) were trained to predict chronological age and get within roughly 3–4 years. The residual, age acceleration, is the signal people care about. It is a small number computed as the difference between two larger, noisy ones, so its measurement error is not negligible.

The residual does track real biology. In the PESA cohort, epigenetic age acceleration associated with subclinical atherosclerosis, with inflammatory markers mediating part of the relationship.4 Age-associated methylation changes and acceleration have been characterized across populations, including comparisons between African American and white middle-aged adults, where some associations replicate across groups and some do not.5 Epigenetic aging trajectories from childhood into early adulthood have been examined against psychotic-like experiences, showing the clocks are being tested well beyond cardiometabolic endpoints.6

The open question is whether the number responds to anything you do. A systematic evaluation of epigenetic aging biomarkers across human longevity intervention trials found that responsiveness varies a great deal by clock and by intervention, with second-generation and pace-of-aging measures behaving differently from first-generation clocks.7 Horvath himself has argued for putting these clocks on trial rather than assuming they are validated surrogate endpoints.8 Newer rate-of-change measures like FraminghamPACE are built to estimate pace directly rather than back it out of a cross-sectional age prediction.9

Our view: treat a clock output as a noisy summary statistic over a genuinely informative measurement. The per-CpG betas are the durable asset. The clock is one model applied to them, and better models will be published. Keep the IDATs.

If you want a single decision rule

Buy the kit only if it returns IDAT files. Run blood, not saliva, if you can, and pair every draw with a CBC with differential. Sample at least three times over a year at the same time of day before drawing any conclusion about trend, because a single pair of points cannot distinguish change from assay noise. Run four or five clocks, not one. If a result concerns you, particularly anything framed as dementia or disease risk, that is a conversation for a clinician who can order confirmatory testing, not something to act on from a consumer report. Community surveys of willingness to take epigenetic dementia risk tests show people want results delivered with clinical support and clear actionability, which consumer kits generally do not provide.10

Questions people also ask

How much does epigenetic testing cost? Consumer kits with a methylation age report run roughly $200–$500 per sample. Running the same EPIC array through a genomics core costs about $150–$300 for the array plus $30–$80 for DNA extraction, and you get the IDATs but no report. Targeted ddPCR clock assays are cheaper per run but not widely sold direct to consumers.

How do you test epigenetics? Extract DNA, treat it with sodium bisulfite to convert unmethylated cytosines, then quantify the converted versus unconverted base at each CpG. The common readouts are Illumina Infinium arrays (~850k–930k sites), whole-genome bisulfite sequencing (every CpG, roughly $1,000+ per sample at useful depth), or targeted PCR at a small panel of sites.

Is a saliva kit as good as blood? No, unless the pipeline corrects for cell composition. Saliva is a variable mix of buccal epithelium and leukocytes, and that ratio drives much of the between-sample variance. Adding cell-type-specific signatures improves buccal-based age predictions considerably.3

Can I recompute my age from someone else’s data format? Only from IDATs or a full beta matrix. A CSV of 20 reported CpGs is not enough to run Horvath’s 353-site clock, and you cannot renormalize what you cannot see.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Lisa M McEwen, Meaghan J Jones, David Tse Shen Lin, et al. Systematic evaluation of DNA methylation age estimation with common preprocessing methods and the Infinium MethylationEPIC BeadChip array. Clinical Epigenetics, 2018. https://doi.org/10.1186/s13148-018-0556-2 ↩

  2. Ilef Hchaichi, Imène Garali, Alina-Madalina Popa, et al. The 12 o’clock assay: an optimized dodecaplex droplet digital PCR assay for robust DNA methylation quantification and epigenetic clock-based age-predictions. Clinical Epigenetics, 2026. https://doi.org/10.1186/s13148-026-02105-0 ↩

  3. Monika Eipel, Felix Mayer, Tanja Arent, et al. Epigenetic age predictions based on buccal swabs are more precise in combination with cell type-specific DNA methylation signatures. Aging, 2016. https://doi.org/10.18632/aging.100972 ↩ ↩2

  4. Fátima Sánchez-Cabo, Valentín Fuster, Juan Carlos Silla-Castro, et al. Subclinical atherosclerosis and accelerated epigenetic age mediated by inflammation: a multi-omics study. European Heart Journal, 2023. https://doi.org/10.1093/eurheartj/ehad361 ↩

  5. Salman M. Tajuddin, Dena G. Hernandez, Brian H. Chen, et al. Novel age-associated DNA methylation changes and epigenetic age acceleration in middle-aged African Americans and whites. Clinical Epigenetics, 2019. https://doi.org/10.1186/s13148-019-0722-1 ↩

  6. Zoe Hart, Anna Großbach, Andrew J Simpkin, et al. The association between epigenetic ageing from childhood to early adulthood and psychotic-like experiences in early adulthood. Psychological Medicine, 2025. https://doi.org/10.1017/s003329172510055x ↩

  7. Raghav Sehgal, Daniel Borrus, Jenel F. Armstrong, et al. Responsiveness of epigenetic aging biomarkers to longevity interventions in humans. Nature Medicine, 2026. https://doi.org/10.1038/s41591-026-04562-9 ↩

  8. Steve Horvath. Putting epigenetic aging clocks on trial. Nature Medicine, 2026. https://doi.org/10.1038/s41591-026-04524-1 ↩

  9. William T Marella, Calen P Ryan, David Corcoran, et al. An epigenetic speedometer to measure Pace of Aging: FraminghamPACE. 2026. https://doi.org/10.64898/2026.07.07.26357388 ↩

  10. Deirdre M. O’Shea, Devi Dhanekula, Swati Kumar, et al. Community perspectives on epigenetic dementia risk testing: Willingness, implementation preferences, and reasons for not testing in midlife and older adults. Alzheimer’s & Dementia, 2026. https://doi.org/10.1002/alz.71094 ↩