In 2020, a research group at the University of California, San Diego published a paper in Cell Systems titled Quantitative Translation of Dog-to-Human Aging by Conserved Remodeling of the DNA Methylome[1]. The paper produced the formula that has replaced the seven-year rule in most modern dog-age calculators:
human_age = 16 × ln(dog_age) + 31
Where ln is the natural logarithm. The formula is logarithmic rather than linear. It says a 1-year-old dog is roughly 31 in human-equivalent age, a 5-year-old is 57, a 10-year-old is 68, and a 15-year-old is 74. It is not a guess or an arithmetic shortcut. It is derived from direct biological measurements of DNA changes that occur at conserved, predictable rates as dogs and humans age.
This article explains how the paper derived that formula, what the formula actually measures, and — maybe most importantly — what it does not measure. The goal is to make the science behind the calculator's backbone formula readable for anyone with high-school-level biology, without losing the nuance that makes the paper rigorous rather than popular.
What is DNA methylation?
DNA is the long chemical polymer that stores genetic information in every cell. It is written in a four-letter alphabet (A, C, G, T) and copied into daughter cells every time a cell divides. The sequence itself is fixed at conception — mutations aside, the A-C-G-T string in a dog's cells does not change over its lifetime.
But the sequence is not the whole story. Overlaid on top of the A-C-G-T string is a second layer of regulatory information called the epigenome. The epigenome determines which parts of the DNA are "on" (being read and turned into proteins) and which parts are "off" (silenced). The epigenome is what makes a liver cell different from a brain cell even though they share the same DNA.
The most-studied part of the epigenome is DNA methylation. Methylation is the addition of a small chemical group — a methyl group, written CH3 — to specific positions on the DNA molecule. Methylation most commonly happens at cytosine (C) bases that sit next to guanine (G) bases, a pattern called a CpG site. The genome has tens of millions of CpG sites, and each can be methylated or unmethylated at any given moment in a cell's life.
Critically, methylation does not change the DNA sequence. It is a reversible chemical decoration layered on top of the sequence. But it has large effects on gene expression — methylated regions are typically silenced, and unmethylated regions are typically active.
Methylation patterns change with age. This is the key fact that makes methylation useful as an aging marker. Specific CpG sites gain methylation as cells age; others lose it. The pattern is not random — it is predictable enough that you can take a DNA sample from a tissue of unknown age and reverse-engineer the chronological age of the donor from the methylation signature alone.
Why methylation makes a good aging clock
Biological aging produces many changes — DNA damage accumulates, telomeres shorten, mitochondria degrade, proteins misfold[10]. Each of these has been studied as a potential aging clock, and each has limitations. DNA damage is noisy and variable. Telomere length correlates with age but with substantial individual variation. Mitochondrial function degrades unevenly across tissues.
Methylation has three properties that make it uniquely useful as a clock:
- High information density. Tens of thousands of CpG sites across the genome age at predictable rates. This gives enough statistical signal to build very accurate predictive models.
- Tissue consistency. Many age-associated methylation sites show similar patterns across tissues. You can measure methylation in a blood sample and infer biological age of tissues you didn't directly sample.
- Cross-species conservation. The biochemistry of methylation is nearly identical across all mammals. The specific sites and rates of methylation change differ by species, but the underlying mechanism is deeply conserved. This is what makes cross-species comparison possible in the first place.
The formal demonstration that methylation is an accurate age predictor came in the early 2010s. Bocklandt et al. (2011) reported that saliva methylation at just two CpG sites could predict human age within about five years[3]. Hannum et al. (2013) built a whole-genome methylation clock for human blood[4]. The paper that defined the field, though, was Horvath (2013).
Horvath's 2013 human methylation clock
Steve Horvath, at UCLA, published in Genome Biology a methylation-based age predictor trained across 51 human tissue types and thousands of donors[2]. The Horvath clock uses 353 specific CpG sites and predicts chronological age with a median error of around 3.6 years across tissues. It works on anything from fetal samples to centenarian blood, and it works on freshly collected samples or samples stored for decades.
The Horvath clock became the reference methylation clock for aging research. It also revealed something unexpected: the predicted "epigenetic age" from the clock sometimes differs from a person's chronological age. A person's biological-age prediction might run 2–3 years ahead of their chronological age, or behind it. That difference — called "epigenetic age acceleration" — turns out to correlate with all-cause mortality, cardiovascular disease, and cancer risk[6]. In other words, the clock measures something real about biological aging, not just calendar time.
The follow-up work was ambitious. Could the same approach work in other species? Could a clock trained on one species be used to predict age in another? These questions set up the canine research.
The dogs-and-wolves predecessor: Thompson et al. 2017
Before Wang et al. (2020), Thompson and colleagues published in Aging a proof-of-concept epigenetic clock for dogs and wolves[5]. They used a targeted methylation array and showed that methylation changed predictably with age in both domestic dogs and gray wolves. The Thompson clock worked — age prediction was accurate within a few years — but it used different CpG sites than the human Horvath clock, and it did not produce a cross-species translation formula.
What Thompson 2017 established: the general methylation-aging principle applies in canids. What it did not produce: a usable dog-to-human age conversion.
Wang et al. 2020 — the paper that built the translation
Wang and colleagues set out to do what Thompson 2017 had not. The research question: can methylation sites that age at conserved rates between dogs and humans be identified, and can those sites be used to build a direct age-translation formula?
The approach had three main steps:
Step 1: Sample. 104 Labrador Retrievers spanning 4 weeks to 16 years of age. Laboratory-raised and companion-animal samples, all from known-age dogs. Labrador was chosen because the breed's aging trajectory is well-characterized and the lifespan (~12 years median) sits in the middle of the canine distribution.
Step 2: Profile methylation. Whole-genome bisulfite sequencing at high coverage, plus targeted arrays. The result: methylation state at tens of thousands of CpG sites per dog, mapped onto a reference canine genome.
Step 3: Compare to humans. 320 humans aged 1 to 103 years, with matched whole-genome methylation data from previously published sources. The researchers aligned the canine methylation data to the human data at orthologous CpG sites — positions in the genome that correspond across species by evolutionary ancestry.
The key finding: a specific set of methylation sites ages at parallel rates between dogs and humans. At these sites, a dog at chronological age X and a human at chronological age Y show the same methylation pattern when X and Y are in a specific mathematical relationship. The researchers then modeled this relationship.
Deriving the formula
Wang et al. tried several functional forms — linear, polynomial, power-law, logarithmic — and evaluated which best fit the conserved-methylation-site data. The winning model was logarithmic:
human_age = 16 × ln(dog_age) + 31
The specific coefficients (16 and 31) came from the best-fit regression on the paired dog-human methylation data at conserved sites. The log relationship reflects the biological reality that dogs age very quickly in the first year or two and much slower later — the rate of methylation change per calendar year is high in puppies and low in seniors, which is exactly what a logarithmic curve describes.
Some worked examples:
| Dog age | ln(dog_age) | 16 × ln(dog_age) | + 31 | Human equivalent |
|---|---|---|---|---|
| 1 year | 0.000 | 0 | 31 | 31 |
| 2 years | 0.693 | 11.1 | 42.1 | ~42 |
| 5 years | 1.609 | 25.7 | 56.7 | ~57 |
| 10 years | 2.303 | 36.8 | 67.8 | ~68 |
| 15 years | 2.708 | 43.3 | 74.3 | ~74 |
Three observations about the curve worth noting:
- A 1-year-old dog maps to 31 human years, not 7. This is the most conceptually important single result. Puppies are biologically young adults, not toddlers.
- Each additional dog-year adds fewer human-years than the last. Going from dog age 1 to 2 adds ~11 human years. Going from dog age 10 to 11 adds ~1.5. Late-life aging, as measured by methylation, genuinely slows per unit of calendar time.
- The curve asymptotes. A 20-year-old dog (extremely rare but documented) maps to ~79 in the formula. The curve flattens hard past middle age.
What the formula measures
The UCSD formula measures epigenetic age — the biological age of the dog's cells as inferred from methylation patterns. This is distinct from chronological age (calendar time since birth) and from life expectancy (years remaining).
The relationship between epigenetic age and chronological age is strong but not perfect. Just as in humans[2], individual dogs of the same chronological age can have somewhat different epigenetic ages — one 7-year-old dog might show methylation patterns consistent with a typical 6.5-year-old, while another shows patterns consistent with 7.5. This individual variation is called epigenetic age acceleration, and it has the same meaning in dogs as in humans: a dog whose epigenetic age runs ahead of its chronological age may be at higher population-level risk of age-related conditions.
The Dog Aging Project[7] is collecting methylation data on a much larger cohort, which will eventually allow the canine version of the "does epigenetic age acceleration predict mortality" question to be answered with the same rigor that Lu et al. (2019) answered it in humans[6].
What the formula does not measure
The UCSD formula does not predict mortality or remaining lifespan. Those questions depend on breed, size, health history, and individual factors that methylation alone cannot capture. Remaining-lifespan estimation belongs to a different statistical question, answered by veterinary life tables[9], not methylation clocks. The Phase 4 companion article how old is my dog in human years walks through how dogage.co layers breed-median percentile on top of the UCSD formula to give the remaining-lifespan context.
The formula does not adjust for breed size. It was derived from Labrador Retrievers — a mid-large breed with ~12-year median lifespan. Very small and very giant breeds age along somewhat different trajectories, and the UCSD formula alone will undershoot or overshoot by a few human-years at the extremes. This is a known limitation noted in the paper itself[1]. The AVMA size-adjusted formula layered alongside UCSD handles the correction.
The formula does not apply to puppies under a few weeks. The log function is undefined at age 0 and produces very large values for very small ages. Practical implementations clamp the formula at age 0.1 years or similar. For puppies, developmental milestones (tooth eruption, growth plate closure, sexual maturation) are more informative than any age-translation formula.
The formula was derived from methylation site biology, not mortality outcomes. It tells you where a dog sits on a biological-age continuum. It does not tell you whether the dog is healthy, whether specific conditions are likely, or how to care for it. Those questions require separate inputs and separate frameworks.
Why the formula works across breeds despite Labrador derivation
A reasonable question: if Wang et al. trained the formula on Labradors, why does it generalize at all to Chihuahuas or Great Danes?
The answer is that the methylation sites identified as "conserved aging sites" between dogs and humans are by construction sites where the underlying biology is species-level rather than breed-level. These are fundamental cellular-aging processes — ones that happen in every dog regardless of breed. The breed-specific layer of aging (size-driven cancer scaling, breed-specific genetic load) operates on top of this common backbone, not instead of it.
Kraus et al. (2013)[8] explicitly decomposed the size-lifespan trade-off in dogs and showed that the signal concentrates in post-adolescent disease incidence rather than intrinsic cellular-aging rate. This is consistent with what the UCSD formula measures: a methylation-based biological-age trajectory that runs similarly across breeds, while the mortality trajectory differs sharply because of breed-specific disease load riding on top.
Where methylation research is going in dogs
Several research threads are active as of 2026:
Larger and more diverse cohorts. The Dog Aging Project[7] has enrolled 47,444 dogs in its longitudinal cohort (per the 2025 cohort-description paper). Methylation profiling on a subset of these dogs will produce cross-breed methylation data at a scale Wang et al. (2020) could not access.
Breed-specific clocks. If enough methylation data accumulates per breed, breed-specific methylation clocks become possible. A Chihuahua clock and a Mastiff clock could capture breed-specific aging-rate differences that a single generic formula misses.
Epigenetic age acceleration and canine mortality. The analogous work to Lu et al. (2019) GrimAge[6] — connecting epigenetic age acceleration to all-cause mortality prediction — is a natural next step in dogs. Longitudinal cohorts with outcome data are the missing ingredient; the Dog Aging Project is built to provide it.
Intervention-response methylation. If an anti-aging intervention (rapamycin, caloric restriction, pharmaceutical candidates) truly slows biological aging, it should show up as slower epigenetic-age progression in treated versus control animals. Methylation clocks give the most sensitive available readout for such comparisons.
What this means for dogage.co's calculator
The UCSD formula is the backbone of the dog age calculator. When you enter your dog's chronological age, the formula runs client-side and returns the methylation-equivalent human age. Three adjustments are layered on top:
- AVMA size-adjusted formula to correct for breed-size effects that UCSD alone misses.
- Breed-median percentile from McMillan et al. (2024) RVC VetCompass data to place the dog in its own breed's lifespan distribution.
- Fortney 2012 life-stage classification for the care-planning bucket.
The combined output is more informative than any single formula. The UCSD number tells you where the dog sits on a biological-age continuum. The size-adjusted value corrects for breed-size. The breed percentile tells you how far through breed-specific lifespan the dog has progressed. The life stage tells you what conversations to have with your veterinarian next. For the practical walkthrough, see how to calculate dog years accurately.
Wang et al. (2020) did not solve every dog-aging question. It did solve the narrow question of how to translate dog chronological age into a biologically-anchored human-equivalent. That is the contribution: a peer-reviewed, measurement-based answer to a question that had been answered for decades with a folk ratio that wasn't anchored to anything. The seven-year rule was never wrong in the sense that 7 is an incorrect number; it was wrong because the real relationship is not a ratio at all. It is a logarithm. Wang et al. measured it.
