Epigenetic DNA Tests: What They Measure and Which Ones Are Worth Running

An epigenetic DNA test almost always means one thing in practice: a measurement of DNA methylation, the addition of a methyl group to cytosine at CpG dinucleotides, across some subset of your genome. It reports, for each site assayed, the fraction of DNA molecules in your sample that carry a methyl mark there. The test does not read your genome sequence, and it does not measure gene expression directly. Consumer versions run $200 to $500 and give you a “biological age” number derived from a few hundred of those sites. Research-grade versions cost more, measure far more sites, and give you data you can reprocess. The distinction that matters is whether you get the raw intensities or reads back, because the summary numbers are the least durable part of the measurement.

What methylation is and why it is measurable

There are roughly 28 million CpG sites in the human genome. Methylation at them is tissue-specific, somewhat heritable, and responsive to environment and age. Methylation in promoter CpG islands is broadly associated with transcriptional repression. Disruption of these patterns is a consistent feature of common disease, particularly cancer, where both focal hypermethylation of tumor suppressor promoters and global hypomethylation are observed.¹ The biology is well established, which answers the “is epigenetics real” question directly: the chemistry is real, the disease associations are real, and the textbook treatment is standard.² What is oversold is the leap from a methylation readout to an actionable statement about your health.

The measurement problem is that a methylated cytosine and an unmethylated cytosine are chemically identical to a sequencer. Every assay works around this. Sodium bisulfite deaminates unmethylated cytosine to uracil (read as thymine) while leaving 5-methylcytosine intact, so methylation becomes a C/T difference you can sequence or hybridize. Enzymatic methyl-seq (NEB’s EM-seq) reaches the same endpoint with TET2 and APOBEC, which avoids the heavy DNA fragmentation bisulfite causes. Affinity methods such as MBD-seq pull down methylated fragments with a methyl-binding domain protein and sequence the enriched pool, giving regional rather than single-base resolution.³ Third-generation platforms skip conversion entirely and infer the modification from the raw current or kinetic signal as the native molecule passes the sensor.⁴

The four platforms, and which one we would choose

The Illumina methylation arrays are the workhorse. The 450K measured about 485,000 CpGs, the EPIC v1 about 850,000, and EPIC v2 roughly 935,000. That is around 3% of the genome’s CpGs, chosen to enrich for promoters, CpG islands, and enhancers. Arrays are cheap and reproducible, and every published methylation clock was trained on them. That is the strongest practical argument for using one.

Whole-genome bisulfite sequencing (WGBS) covers nearly every CpG at single-base resolution and does not depend on probe design choices made years ago. Sequencing-based epigenomics has clear advantages over hybridization. There is no probe bias and no fixed content, the dynamic range is better, and phasing reads reveals allele-specific methylation.⁵ The cost is depth. You need something like 25 to 30x to estimate a per-site beta value with usable precision, because at depth 10 the binomial standard error on a beta near 0.5 is about 0.16. EM-seq is the version we would run, since it preserves longer fragments and gives more even coverage than bisulfite at equal depth.

Nanopore sequencing is the most interesting option if you are already sequencing your genome. A single run yields genotype, structural variants, phasing, and 5mC plus 5hmC calls from the same reads, with no conversion chemistry and no separate library.⁴ Read lengths of tens of kilobases let you assign methylation to a specific haplotype, which arrays cannot do. Accuracy at individual CpGs is lower than bisulfite sequencing, so region-level summaries are more trustworthy than single sites.

Our recommendation: if you want a number you can compare against published literature, run an EPIC array. If you want data you will still be reanalyzing in five years, run EM-seq or nanopore and accept that most clock implementations will require imputing array probes from your coverage. The broader case for sequencing over arrays in epigenetics is that the assay is not frozen at the moment of purchase.⁶

The files you get and how to process them

From an array, the raw output is a pair of IDAT files per sample (_Grn.idat and _Red.idat) plus a sample sheet. Insist on these. If a vendor only returns a PDF, you have bought an opinion. You have not bought a measurement.

In R, we would start with sesame rather than minfi for EPIC v2, because its masking of unreliable probes is better maintained:

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

The QCDPB preparation chain applies channel inference, dye bias correction, background subtraction with noob, and probe masking. Then drop probes with detection p-value above 0.01, probes overlapping common SNPs, and the cross-reactive probes on the published exclusion lists. Convert to M-values, log2(beta / (1 - beta)), for any regression. Beta values are heteroscedastic near 0 and 1.

From sequencing, align with bwa-meth and call with MethylDackel:

bwameth.py --reference hg38.fa R1.fq.gz R2.fq.gz | samtools sort -o s.bam
MethylDackel extract --mergeContext --minDepth 10 --OT 5,0,5,0 hg38.fa s.bam

The --OT/--OB trimming arguments remove the M-bias at read ends that every bisulfite library shows. Check it first with MethylDackel mbias. Always include a spike-in: unmethylated lambda DNA to measure conversion efficiency (you want above 99%) and CpG-methylated pUC19 to measure false conversion. A conversion rate of 97% inflates apparent methylation across millions of sites and will quietly shift every downstream estimate. For nanopore, dorado basecaller --modified-bases 5mCG_5hmCG followed by modkit pileup gives you a bedMethyl file with per-site counts and can be split by haplotype with --partition-tag HP.

Clocks, and what they are estimating

Epigenetic age predictors are penalized regressions on array beta values. Horvath’s multi-tissue clock uses 353 CpGs, Hannum’s blood clock 71, and PhenoAge 513. DunedinPACE uses 173 sites trained on rate of change in physiological markers rather than chronological age. These are regression outputs. They are not measurements of a biological quantity called age. Two failure modes dominate.

The first is cell composition. Whole blood is a mixture, and neutrophils, CD8 T cells, and B cells have sharply different methylomes. A shift in your differential count from an infection the week of the draw moves clock estimates. Run Houseman-style deconvolution (estimateCellCounts2 with the FlowSorted.Blood.EPIC reference) and report the cell fractions alongside any age estimate. If someone hands you an epigenetic age without cell proportions, treat it as uninterpretable.

The second is genetics. A substantial fraction of variance in methylation at individual CpGs is under genetic control, and family studies show widespread heritable influence on methylation state.⁷ Part of the number you get back is your genotype, fixed at conception and not responsive to anything you do. This is why clock changes after an intervention should be read against test-retest noise, which for single-timepoint array clocks is often several years of apparent age. Longitudinal comparison on the same platform, tissue, and lab is the only version of this measurement we would act on. Here “act on” means generating a hypothesis rather than changing anything medical. Interpretation of any result suggesting disease belongs with a clinician.

Questions people also ask

How much does an epigenetic test cost?

Direct-to-consumer methylation age tests run roughly $200 to $500 for a saliva or blood spot and an EPIC array. Research-grade WGBS or EM-seq at 30x costs considerably more, driven almost entirely by sequencing depth rather than library prep. Formal analyses of sequencing economics consistently find that reported per-sample costs omit analysis, storage, and interpretation. Those items can equal or exceed the wet-lab line item.⁸

What does an epigenetic test do?

It reports the fraction of your DNA molecules carrying a methyl group at each assayed CpG site, in the specific tissue sampled. From that matrix, software estimates cell-type proportions, age predictors, and sometimes smoking or metabolic surrogates. It does not tell you which genes are currently being transcribed, which requires RNA sequencing.

Is epigenetics real or fake?

The molecular biology is settled and the assays are quantitative and reproducible.² What is contested is the commercial layer: the claim that a single clock reading reflects your health trajectory, and that specific products move it. Treat the chemistry as solid and the interpretation as preliminary.

What diseases are linked to epigenetics?

Cancer is the best characterized, with recurrent promoter hypermethylation and global hypomethylation across tumor types, and genome-scale studies have extended methylation associations to common metabolic and autoimmune conditions.¹ Imprinting disorders such as Prader-Willi and Angelman arise from defects in parent-of-origin methylation. Association in these studies does not establish causal direction, and none of it constitutes a diagnosis from a consumer test.

Footnotes

1. Andrew P. Feinberg. Genome-scale approaches to the epigenetics of common human disease. Virchows Archiv, 2009. https://doi.org/10.1007/s00428-009-0847-2 ↩ ↩²

2. Carsten Carlberg, Ferdinand Molnár. Human Epigenetics: How Science Works. 2019. https://doi.org/10.1007/978-3-030-22907-8 ↩ ↩²

3. David Serre, Byron H. Lee, Angela H. Ting. MBD-isolated Genome Sequencing provides a high-throughput and comprehensive survey of DNA methylation in the human genome. Nucleic Acids Research, 2009. https://doi.org/10.1093/nar/gkp992 ↩

4. Bethany Searle, Markus Müller, Thomas Carell, et al. Third‐Generation Sequencing of Epigenetic DNA. Angewandte Chemie, 2023. https://doi.org/10.1002/ange.202215704 ↩ ↩²

5. P. J. Hurd, C. J. Nelson. Advantages of next-generation sequencing versus the microarray in epigenetic research. Briefings in Functional Genomics and Proteomics, 2009. https://doi.org/10.1093/bfgp/elp013 ↩

6. Chee Seng Ku, Nasheen Naidoo, Mengchu Wu, et al. Studying the epigenome using next generation sequencing. Journal of Medical Genetics, 2011. https://doi.org/10.1136/jmedgenet-2011-100242 ↩

7. Jason Gertz, Katherine E. Varley, Timothy E. Reddy, et al. Analysis of DNA Methylation in a Three-Generation Family Reveals Widespread Genetic Influence on Epigenetic Regulation. PLoS Genetics, 2011. https://doi.org/10.1371/journal.pgen.1002228 ↩

8. Kurt Christensen, Dmitry Dukhovny, Uwe Siebert, et al. Assessing the Costs and Cost-Effectiveness of Genomic Sequencing. Journal of Personalized Medicine, 2015. https://doi.org/10.3390/jpm5040470 ↩