How can we objectively measure our level of aging?
Chronological age (on ID cards) clearly fails to fully reflect one’s actual physical health. A better alternative may lie in the chemical marks on DNA that record our life journey. Indeed, scientists have developed epigenetic clocks—tools that assess biological age by decoding these DNA markers.
Yet a problem emerged: the field once descended into chaos with conflicting conclusions. When evaluating the same anti-aging supplement, different clocks might yield opposite results—some claiming you’ve grown 5 years younger, others that you’ve aged 3 years. So… who should we trust? Is there a recognized, reliable standard?
Fear not. A recent study published in GeroScience aims to resolve this chaos. It not only creates a more reliable new clock but also reveals that our "innate" aging rate can be quantified—and used to predict future natural lifespan.
01 The Trouble with "Unregulated Growth"
Why do different epigenetic clocks give conflicting conclusions about the same anti-aging intervention? The root cause lies in the field’s current "unregulated growth" phase[2].
Figure note: Over the past decade, different research teams have developed distinct clocks for varied goals
| Year | Clock Name | Features |
|---|---|---|
| 2013 | Horvath's Pan-Tissue Clock | 353 CpGs, applicable to multiple tissues |
| - | Hannum's Clock | 71 CpGs, for whole blood |
| - | Horvath's Skin & Blood Clock | 391 CpGs, measured in ex vivo studies, applicable to multiple tissues |
| 2018 | PhenoAge (Levine) | 513 CpGs, predicts phenotypic age and mortality risk |
| - | GrimAge (Horvath) | 1030 CpGs, applicable to multiple tissues, predicts lifespan and healthspan |
While these achievements are impressive, inconsistent evaluation standards sowed the seeds of future contradictions:
-
Heterogeneity in design: Each clock has a unique design philosophy—differences in training data (healthy vs. diseased populations), selected DNA sites, and underlying statistical algorithms. It’s like asking economists and sociologists to analyze the same social issue: varying perspectives lead to divergent conclusions.
Figure note: Different designs result in distinct clocks -
Insensitivity to interventions: Most early clocks are accurate at estimating chronological age but struggle to answer, "Did this drug/intervention make me younger?" They often fail to detect stable, accurate signals of subtle biological changes caused by anti-aging interventions.
Figure note: For calorie restriction (CR group), PhenoAge and GrimAge detected no statistically significant differences compared to the control group (AL group)[3]
Have we got our priorities wrong? We’ve fixated on clocks’ accuracy in predicting age, while ignoring their more critical role as "sensitive tools for evaluating interventions." After all, what use is a tool that can’t reliably answer, "What effectively slows aging?"
02 Taming All Clocks in Two Steps
Scientists realized: Instead of competing to create "better" single clocks, it’s wiser to combine the strengths of existing clocks into a more comprehensive "ensemble" clock.
First, a universal benchmark was needed—and that’s where MethylGauge came in. This standardized "anti-aging intervention benchmark library" abandons simple age-sequenced data, instead compiling 211 datasets from real intervention experiments with clear biological outcomes.
Figure note: The library includes anti-aging interventions (e.g., calorie restriction, cellular reprogramming, exercise, rapamycin) and pro-aging stressors (e.g., progeria genetic models, high-fat diet, sleep deprivation)—with known expected biological effects for each experiment
Using the MethylGauge dataset, the team developed the new clock—EnsembleAge—available in two versions:
No.1 The "Pro Max" Version—EnsembleAge.Dynamic
This core version is a "thinking" clock. When evaluating a new intervention, it automatically selects clock models most responsive to that intervention, adopts only their most sensitive predictions, and takes the median as the final result.
Figure note: The dynamic adaptive system converts weak intervention signals from traditional clocks (top) into an ensemble model (bottom) that amplifies rejuvenation (purple) and stress (blue) effects for precise assessment
Figure note: The dynamic adaptive system converts weak intervention signals from traditional clocks (top) into an ensemble model (bottom) that amplifies rejuvenation (purple) and stress (blue) effects for precise assessment
No.2 The Standard Version—EnsembleAge.Static
To simplify complex calculations, researchers also developed a "static version." This pre-trained single model predicts the complex results of the dynamic version—concise, efficient, and ideal for initial screening of large sample sizes.
Figure note: The y-axis (static version) closely fits and predicts the x-axis (dynamic version) results
Figure note: The y-axis (static version) closely fits and predicts the x-axis (dynamic version) results
In short, EnsembleAge’s core principle is: Value is proven on a unified benchmark, not self-declared. This ensures stable, reliable evaluations and solves the old problem of clocks "speaking different languages" about interventions. Now, let’s see how EnsembleAge performs with the support of MethylGauge!
03 All-Round Performance?
Predicting Natural Lifespan
One of EnsembleAge’s most impressive feats is its ability to predict mice’s natural lifespan—a long-standing challenge for epigenetic clocks. Most clocks are designed to "match" chronological age or predict short-term health risks, not the natural endpoint of healthy individuals.
Results showed: By analyzing samples from mice in early life (only 20 days after birth), EnsembleAge calculated their innate aging rate—and this rate correlated strongly negatively with their future natural lifespan.
Figure notes (from left to right): Age measured in multiple tissues (R=0.98, p<2.2e-16); Blood samples from old mice vs. remaining lifespan (R=-0.45, p=0.00027); Tail samples from newborn mice vs. remaining lifespan (R=-0.46, p=0.0013)
In other words, mice born with a faster aging rate were destined to die earlier.
Accurately Detecting Subtle Biological Signals
Beyond predicting long-term lifespan, EnsembleAge also excels at evaluating short-term, subtle changes.
Whether in a Huntington’s disease model (where many traditional clocks fail to detect pro-aging effects) or under rapamycin intervention (where many clocks can’t confirm rejuvenation), EnsembleAge clearly identified significant pro-aging/rejuvenating effects.
Whether in a Huntington’s disease model (where many traditional clocks fail to detect pro-aging effects) or under rapamycin intervention (where many clocks can’t confirm rejuvenation), EnsembleAge clearly identified significant pro-aging/rejuvenating effects.
Figure note (left): Huntington’s disease model (Ensemble Z=2.99); Figure note (right): Rapamycin intervention (Ensemble Z=-2.96). Each colored bar represents one clock’s judgment—bars extend outward for aging acceleration (left) and inward for rejuvenation (right)
Breaking the Species Barrier
The team went further: To bridge the critical gap from "mouse experiments" to "human applications," they developed a human-mouse universal EnsembleAge. By analyzing conserved DNA methylation sites between the two species, it establishes a unified standard for aging assessment—accelerating the clinical translation of anti-aging research.
Figure note: EnsembleAge outperforms all original clocks in overall responsiveness (left), rejuvenation response efficiency (middle), and stress response efficiency (right)—even the human-mouse universal version (red box) performs excellently
EnsembleAge’s prowess as a top scientific tool is clear. But digging deeper into its background reveals a clear industrial strategy behind this breakthrough.
Nearly all core authors of this paper come from Altos Labs—a company that has risen to prominence in recent years.
Figure note: Altos Labs’ internal work environment
04 More Than Just a Clock
Altos Labs is no ordinary startup. Founded by Russian billionaire Yuri Milner, with investors including Amazon founder Jeff Bezos, it announced $3 billion in committed funding at launch—an unprecedented sum for a biotech startup at the time.
Its team is a "Nobel Prize-level dream team," including Shinya Yamanaka (father of reprogramming technology) and Jennifer Doudna (CRISPR pioneer). In short, it has assembled some of the world’s top researchers and most abundant startup capital.
With this context, Altos Labs’ core research focus—epigenetic reprogramming (using "Yamanaka factors" to reverse cellular aging)—makes sense. To prove this technology’s effectiveness, a highly reliable evaluation tool is essential: Enter EnsembleAge. It is not just a scientific tool, but a key link in Altos’ business loop.
Yet challenges remain:
- Developing a human MethylGauge: While the "human-mouse universal clock" is a major step forward, serving human applications requires a human intervention database of equal quality—supported by extremely expensive and lengthy clinical trials.
- From precise measurement to precise guidance: Future aging clocks should not just report a biological age number, but advise: "Your aging mainly affects the XX pathway in your XX organ; we recommend XX targeted intervention."
EnsembleAge has laid a solid foundation, but this will require more massive multi-omics data and complex algorithms.
These challenges are also opportunities. We may be very close to the era of true "precision anti-aging."
References
[1] Haghani A, Lu AT, Yan Q, Belmonte JCI, Reddy P, Cheng V, Yang XW, Wang N, Mozhui K, Murach K, Ocampo A, Williams RW, Jucker M, Bergmann C, Poganik JR, Zhang B, Gladyshev VN, Horvath S. EnsembleAge: enhancing epigenetic age assessment with a multi-clock framework. Geroscience. 2025 Aug 6. doi: 10.1007/s11357-025-01808-1. Epub ahead of print. PMID: 40768061.
[2] Topart C, Werner E, Arimondo PB. Wandering along the epigenetic timeline. Clin Epigenetics. 2020 Jul 2;12(1):97. doi: 10.1186/s13148-020-00893-7. PMID: 32616071; PMCID: PMC7330981.
[3] Waziry R, Ryan CP, Corcoran DL, Huffman KM, Kobor MS, Kothari M, Graf GH, Kraus VB, Kraus WE, Lin DTS, Pieper CF, Ramaker ME, Bhapkar M, Das SK, Ferrucci L, Hastings WJ, Kebbe M, Parker DC, Racette SB, Shalev I, Schilling B, Belsky DW. Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Nat Aging. 2023 Mar;3(3):248-257. doi: 10.1038/s43587-022-00357-y. Epub 2023 Feb 9. Erratum in: Nat Aging. 2023 Jun;3(6):753. doi: 10.1038/s43587-023-00432-y. PMID: 37118425; PMCID: PMC10148951.
[2] Topart C, Werner E, Arimondo PB. Wandering along the epigenetic timeline. Clin Epigenetics. 2020 Jul 2;12(1):97. doi: 10.1186/s13148-020-00893-7. PMID: 32616071; PMCID: PMC7330981.
[3] Waziry R, Ryan CP, Corcoran DL, Huffman KM, Kobor MS, Kothari M, Graf GH, Kraus VB, Kraus WE, Lin DTS, Pieper CF, Ramaker ME, Bhapkar M, Das SK, Ferrucci L, Hastings WJ, Kebbe M, Parker DC, Racette SB, Shalev I, Schilling B, Belsky DW. Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Nat Aging. 2023 Mar;3(3):248-257. doi: 10.1038/s43587-022-00357-y. Epub 2023 Feb 9. Erratum in: Nat Aging. 2023 Jun;3(6):753. doi: 10.1038/s43587-023-00432-y. PMID: 37118425; PMCID: PMC10148951.