Epigenetic clocks compared - Horvath, GrimAge, DunedinPACE

Epigenetic clocks are often discussed as if they were different versions of one number. That is wrong. Horvath, PhenoAge, GrimAge and DunedinPACE answer different questions.

Any product, article or report that uses them has to start with a basic distinction: is the clock estimating state, risk, biological age or speed of change?

Context

The first well-known generation of epigenetic clocks focused on age reconstruction. Horvath's multi-tissue clock showed that DNA methylation patterns across many tissues carry a strong age-related signal. The paper DNA methylation age of human tissues and cell types (https://pubmed.ncbi.nlm.nih.gov/24138928/) became a reference point for the field.

The question then shifted. The field moved beyond estimating chronological age. Models such as PhenoAge and GrimAge were tied more directly to health-related endpoints. DunedinPoAm and DunedinPACE shifted the frame again: not only "how old", but "how fast".

That distinction matters for ARES. A state score can be interesting. For navigation, speed is often more useful because it makes change visible.

Four clock frames

Horvath: age reconstruction

Horvath answers a direct question: how old does a tissue look based on its methylation pattern? The strength is a robust age signal across tissues. The product limitation is also clear: a biological age delta is easy to dramatize and harder to use responsibly.

PhenoAge: closer to clinical risk

PhenoAge, described by Levine et al. (https://pubmed.ncbi.nlm.nih.gov/29676998/), connects methylation to a phenotypic age construct that is closer to mortality- and healthspan-related endpoints. That makes the score more clinically interesting than pure age reconstruction, while still leaving it dependent on model, population and calibration.

GrimAge: endpoint proximity

GrimAge, described by Lu et al. (https://pubmed.ncbi.nlm.nih.gov/30669119/), is often discussed because of its association with lifespan- and healthspan-related endpoints. The name sounds severe, but the model remains a statistical construct. Product communication around GrimAge needs special restraint because users can easily read the score as fate.

DunedinPACE: velocity

DunedinPACE, described in eLife 2022 (https://pubmed.ncbi.nlm.nih.gov/35029144/), models the pace of biological aging. This is the most natural fit for Bio-Velocity: a pace signal can be interpreted alongside sleep, HRV, training load, inflammation and lipids.

Comparison

| Clock | Primary question | Strength | Product risk | |---|---|---|---| | Horvath | How old does tissue look? | Strong age signal across tissues | Biological age is dramatized | | PhenoAge | How close is the score to phenotypic risk? | Closer to health endpoints | Risk is read as diagnosis | | GrimAge | Which endpoint-related signal appears? | Strong healthspan/lifespan associations | Score sounds more final than it is | | DunedinPACE | How fast is the system changing? | Dynamics for navigation | Small changes are overinterpreted |

Why pace matters more for ARES

ARES should not become a score gallery. The relevant question is not which number looks most impressive. The relevant question is which signal helps explain the course of the system.

A pace signal fits that logic better. It can sit next to other time-dependent signals:

• sleep architecture over several weeks
• HRV baseline and stress periods
• inflammation markers across quarters
• lipid and glucose development
• training load and subjective recovery

When several layers move in the same direction, a hypothesis appears. When they contradict each other, a review question appears. Both are more useful than an isolated score.

Quality requirements for content and pipeline

Every autonomous blog or social pipeline that describes epigenetic clocks needs hard rules:

1. Name the clock type. Horvath, GrimAge and DunedinPACE are not interchangeable. 2. State proxy status. A score is model output, not the person. 3. Limit single-value claims. One point is not a trajectory. 4. Enforce claim safety. No diagnosis, treatment claim or outcome promise. 5. Preserve metadata. Lab, platform, sample context and comparability matter.

This is where marketing automation has to become professional. A fast article is not enough. The worker has to understand which clock it is describing and which conclusions are not allowed.

Risks

Epigenetic clocks are especially vulnerable to overclaiming because they sound scientific and translate easily into consumer stories.

Common failures include:

• treating biological age as a hard truth
• selling a change between two tests as proof that an intervention worked
• reading study associations as individual prognosis
• comparing different assays and clock generations directly
• turning proxy signals into product promises

For ARES, the rule is simple: clocks are context signals. They can enter the navigation layer, but they do not steer alone.

Key takeaways

• Horvath, PhenoAge, GrimAge and DunedinPACE answer different questions.
• Pace is often more useful for navigation than an isolated biological age score.
• Every clock remains model output with limits.
• ARES should fuse clock signals with other time series and enforce claim safety.

Disclaimer

This document is for education and scientific context only. It is not diagnosis, treatment advice or self-medication guidance. Epigenetic testing needs qualified clinical context whenever health decisions depend on it.

Sources

• Horvath S. DNA methylation age of human tissues and cell types. Genome Biology. 2013. DOI: 10.1186/gb-2013-14-10-r115. PMID: 24138928.
• Levine ME et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging. 2018. DOI: 10.18632/aging.101414. PMID: 29676998.
• Lu AT et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging. 2019. DOI: 10.18632/aging.101684. PMID: 30669119.
• Belsky DW et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. 2022. DOI: 10.7554/eLife.73420. PMID: 35029144.

<!-- tags: Epigenetics, Clocks, Horvath, GrimAge, DunedinPACE -->