Epigenetic clocks estimate how fast a body is ageing by reading chemical tags on DNA, and the first ones were applied to men and women alike. Research since then shows that the two sexes do not age along the same path, so one averaged score can blur differences that matter. A clock retaken every few months also produces a record far more revealing than a one-off ancestry test. That raises a second question: who gets to keep it?
Medicine has averaged before
The habit of treating one body as the default is not new. In 1977 the US Food and Drug Administration issued a guideline that excluded women of childbearing potential from early-phase drug studies. A replacement guideline in July 1993 reversed it, and the NIH Revitalization Act, signed on 10 June 1993, required NIH-funded researchers to include women in human subject research. Mazure and Jones, 2015 trace that history.
The gap left a measurable mark. A 2001 US General Accounting Office report found that eight of the ten prescription drugs withdrawn from the US market since January 1997 posed greater health risks for women than for men. In 2013 the FDA told manufacturers to lower the recommended dose of the insomnia drug zolpidem for women from 10 mg to 5 mg, because women eliminate it from the body more slowly than men.
A 2020 review by Zucker and Prendergast looked at 86 FDA-approved drugs. Of those with higher blood levels in women, 96% were linked to more adverse reactions in women, and across 59 drugs with identifiable reactions, the sex difference in blood levels predicted the direction of the difference in adverse reactions 88% of the time. Their conclusion was that equal doses for women and men risk overmedicating women. When a field averages two populations, it tends to calibrate to whichever one it studied first.
What the ageing literature shows
Steve Horvath’s 2013 clock estimated age from 353 CpG sites, positions on the genome where methyl groups attach, using about 8,000 samples covering 51 healthy tissues and cell types. It was built to work across tissues. Later studies applied the same clock to both sexes and looked for differences. Horvath’s own 2016 analysis found that men have higher epigenetic ageing rates than women in blood, saliva and brain tissue, a result a 2021 review in Aging Cell summarised as slightly faster ageing in men on the Horvath, Hannum and Zbieć-Piekarska clocks.
That is only half the picture, because women live longer yet are frailer. Hägg and Jylhävä’s 2021 eLife review describes a paradox: women generally live longer, consistent with lower biological ages on molecular biomarkers, but have worse health at the end of life. A score that reads “younger” does not mean “healthier”, and the meaning of the same number may differ by sex.
The strongest evidence that the sexes age differently at the molecular level comes from methylation studies:
| Finding | Study | Why it matters for a pooled clock |
|---|---|---|
| In 2,586 adults aged 18 to 87, replicated in 4,450 more, 52 autosomal and 597 X-linked CpG sites showed different age trajectories by sex. Most of the differences sit on the X chromosome. | McCartney et al., Genome Medicine, 2019 | The region where the sexes diverge most is one that analyses often leave out. |
| Sex chromosome data are “frequently excluded” from large-scale analyses because of analytical complexity. | Inkster et al., Epigenetics & Chromatin, 2023 | A pooled model can miss the signal without anyone deciding to. |
| A meta-analysis of four blood datasets found 43% of sex-associated probes change with age, and the number of probes with rising variability is 15 times higher in males. | Yusipov et al., Aging, 2020 | Male and female ageing differs in spread as well as in average. |
| Earlier menopause, bilateral oophorectomy and a longer time since menopause were linked to higher epigenetic age acceleration in blood. | Levine et al., PNAS, 2016 | A hormonal event specific to women moves the clock. |
One caution runs against the simple story. In the same Levine study, epigenetic age acceleration in saliva and cheek cells did not relate to age at menopause, while blood did. Tissue matters as much as sex, and a saliva test cannot be assumed to behave like a blood test.
So the evidence supports building sex into ageing clocks. It does not yet show that separate clocks predict health outcomes better than a pooled clock with sex as a covariate. Blockleaders found no independent, peer-reviewed head-to-head comparison on health outcomes.
One company’s answer
Sam Thompson, founder of DEPH Network, whose DEPH Labs group builds the clocks behind the UmmuHub and ENGNE apps, is among those betting on the split. Describing the approach on Blockleaders’ Cognitio show, he said: “we’re separating the averages, we’re making it sex specific.” DEPH Labs’ UmmuAge clock for women was, by the company’s account, trained on 1,521 women and validated on 8,436 women across 41 CpG sites, and a companion male clock, AlphaAge, covers 285 sites. Those figures are the company’s own, and this article has not independently checked them.
The record outlives the test
A clock result sounds harmless. The underlying file is not. Philibert and colleagues showed that methylation array data can reveal genotype at 1,069 informative loci, along with alcohol and smoking consumption. The same file can identify a person and describe their habits. Someone who retests every quarter to see whether a diet or hormone therapy is working builds a time series of their own physiology.
The consumer genetics industry has already shown what can go wrong with custody. In 2023 attackers used passwords leaked elsewhere to log into about 14,000 23andMe accounts, then reached other users through the company’s relative-matching feature, and about 6.9 million people had their data stolen. In March 2025 the company filed for Chapter 11 bankruptcy, raising concern over the personal and genetic information of more than 15 million consumers. Anne Wojcicki’s nonprofit, TTAM Research Institute, was named successful bidder at $305 million, with Regeneron as the rival bidder. In July 2026 Delaware’s attorney general and 42 others announced a settlement that included $150 million in allowed claims for states, though about $18 million was actually paid from bankruptcy funds. The sale to TTAM carried added security requirements and continuing deletion rights. Our earlier piece on 23andMe’s collapse covers the detail.
Regulation does not fill the gap. Most direct-to-consumer genetic testing companies are not HIPAA “covered entities”, so their customers rely on the company’s own privacy policy and a patchwork of state laws.
Who should hold it
There are three broad models for custody, each with a different failure mode.
| Model | Who holds the data | Main exposure |
|---|---|---|
| Company database | The testing company, on its own or a cloud server | Breach, bankruptcy sale, policy changes. Outside HIPAA for most consumer tests |
| Clinician-held | A provider or health plan acting under HIPAA | Strong rules, but covers only care delivered through that channel |
| User-held, computed in sealed hardware | The user holds the keys; analysis runs inside a protected enclave | Trust shifts from the company to the chip and its maker |
DEPH Network, which Thompson founded, is building toward the third model. Its site says health data is “never on the chain” and stays encrypted off-chain, with the chain holding only the record of access and rewards. The compute side relies on Secret Network’s hardware enclaves, where, per its documentation, the state inside the enclave is never revealed. Hardware trust has limits. In 2018 researchers showed that the Foreshadow attack could read an Intel SGX enclave’s entire memory contents. That makes the model a reduction in exposure rather than an end to it. Part 3 of this series looks at the AI coach built on that stack.
What to ask before you test
Whichever model a provider uses, five questions are worth putting to them:
- Which sex and tissue was the clock built and validated on, and is that published in a peer-reviewed paper or only on the company’s site?
- Who can read the raw methylation file, not only the summary score?
- What happens to the data if the company is sold or goes bankrupt?
- Can you delete it, and does deletion cover the raw file and any derived models?
- Is the service covered by HIPAA, or by a privacy policy the company can change?
The first question applies to the science and the other four to the custody. Both have to hold up. A clock tuned to the right sex but kept in the wrong hands solves one problem and creates another.
This article is part of Blockleaders’ coverage of Cognitio, the Blockleaders show hosted by Justin Roberti, from the episode “The AI Secret to Better Health: DEPH Network on Secret3” with Lisa Loud, Sam Thompson and Ian Spiegel. Watch the full episode.
Sources
Every source below was read during research for this rewrite. Where a publisher site blocked automated access, the abstract or full text was read through Europe PMC, noted in brackets. Full list:
- Mazure CM, Jones DP. Twenty years and still counting: including women as participants and studying sex and gender in biomedical research. BMC Women’s Health, 2015. https://pmc.ncbi.nlm.nih.gov/articles/PMC4624369/ (read via Europe PMC full text: https://www.ebi.ac.uk/europepmc/webservices/rest/PMC4624369/fullTextXML)
- US Government Accountability Office (then General Accounting Office). Drug Safety: Most Drugs Withdrawn in Recent Years Had Greater Health Risks for Women, GAO-01-286R, January 2001. https://www.gao.gov/products/gao-01-286r
- FDA Drug Safety Communication, 10 January 2013, zolpidem dosing. https://www.fda.gov/files/drugs/published/Drug-Safety-Communication–Risk-of-next-morning-impairment-after-use-of-insomnia-drugs–FDA-requires-lower-recommended-doses-for-certain-drugs-containing-zolpidem-%28Ambien–Ambien-CR–Edluar–and-Zolpimist%29.pdf
- Zucker I, Prendergast BJ. Sex differences in pharmacokinetics predict adverse drug reactions in women. Biology of Sex Differences, 2020. https://bsd.biomedcentral.com/articles/10.1186/s13293-020-00308-5 (abstract read via Europe PMC)
- Horvath S. DNA methylation age of human tissues and cell types. Genome Biology, 2013. https://pubmed.ncbi.nlm.nih.gov/24138928/ (abstract read via Europe PMC)
- Horvath S, et al. An epigenetic clock analysis of race/ethnicity, sex, and coronary heart disease. Genome Biology, 2016. https://genomebiology.biomedcentral.com/articles/10.1186/s13059-016-1030-0 (abstract read via Europe PMC)
- Simpson DJ, Chandra T. Epigenetic age prediction. Aging Cell, 2021. https://pmc.ncbi.nlm.nih.gov/articles/PMC8441394/
- Hägg S, Jylhävä J. Sex differences in biological aging with a focus on human studies. eLife, 2021. https://elifesciences.org/articles/63425 (abstract read via Europe PMC)
- McCartney DL, et al. An epigenome-wide association study of sex-specific chronological ageing. Genome Medicine, 2019. https://link.springer.com/article/10.1186/s13073-019-0693-z (abstract read via Europe PMC)
- Inkster AM, et al. Who’s afraid of the X? Incorporating the X and Y chromosomes into the analysis of DNA methylation array data. Epigenetics & Chromatin, 2023. https://pmc.ncbi.nlm.nih.gov/articles/PMC9825011/ (abstract read via Europe PMC)
- Yusipov I, et al. Age-related DNA methylation changes are sex-specific: a comprehensive assessment. Aging (Albany NY), 2020. https://pmc.ncbi.nlm.nih.gov/articles/PMC7762479/ (abstract read via Europe PMC)
- Levine ME, et al. Menopause accelerates biological aging. PNAS, 2016. https://www.pnas.org/doi/full/10.1073/pnas.1604558113 (abstract read via Europe PMC, PMID 27457926)
- Philibert RA, et al. Methylation array data can simultaneously identify individuals and convey protected health information: an unrecognized ethical concern. Clinical Epigenetics, 2014. https://clinicalepigeneticsjournal.biomedcentral.com/articles/10.1186/1868-7083-6-28 (abstract read via Europe PMC)
- HIPAA Journal. 6.9 million 23andMe users affected by data breach. https://www.hipaajournal.com/6-9-million-23andme-users-affected-by-data-breach/
- ABC News. 23andMe has filed for bankruptcy. What could happen to users’ genetic data? https://abcnews.com/Business/23andme-filed-bankruptcy-happen-users-genetic-data/story?id=120090459
- Quinn Emanuel. Win for Anne Wojcicki and TTAM Research Institute in the 23andMe bankruptcy auction. https://www.quinnemanuel.com/the-firm/publications/quinn-emanuel-secures-an-unprecedented-win-for-anne-wojcicki-and-ttam-research-institute-in-23andme-bankruptcy-auction/
- Delaware Department of Justice. AG Jennings announces multistate settlement of bankruptcy claims against 23andMe, 14 July 2026. https://news.delaware.gov/2026/07/14/ag-jennings-announces-multistate-settlement-of-bankruptcy-claims-against-23andme-over-genetic-data-breach/
- AccountableHQ. HIPAA and direct-to-consumer genetic testing: what applies and what doesn’t. https://www.accountablehq.com/post/hipaa-and-direct-to-consumer-genetic-testing-what-applies-and-what-doesn-t
- UmmuHub. The Science (UmmuAge cohorts and CpG sites). https://ummuhub.com/pages/the-science
- DEPH Labs (AlphaAge: 285 CpG sites, 2,097 men validated, company-reported). https://dephlabs.com/
- DEPH Network. https://dephnetwork.com/
- Secret Network documentation, Intel SGX. https://docs.scrt.network/secret-network-documentation/introduction/secret-network-techstack/privacy-technology/intel-sgx
- Foreshadow attack on Intel SGX enclaves (2018), project site. https://foreshadowattack.eu/
- Cognitio episode transcript (Whisper, verbatim), 32:42 to 32:52, quote used once. Internal file: TRANSCRIPT_WHISPER_VERBATIM.txt
