In the research field of aging biology, the epigenetic clock has developed from a novel concept to an indispensable and powerful tool. However, with the emergence of new generation clocks from pioneering models such as Horvath and Hannum, to PhenoAge and GrimAge, which are closely related to health risks, and to DunedinPACE, which dynamically monitors the aging rate, researchers are facing a choice dilemma.

This paper aims to provide a systematic selection guide to solve this problem. It will first sort out the intergenerational differences and core mission of the first, second and third generation clocks, and clarify the fundamental change of their design philosophy. Then it discusses how to accurately match the models of different generations according to the specific research objectives. Thirdly, a data-driven practical model selection framework is proposed, and finally the decision boundary of when to adopt ready-made models and when to build customized clocks is defined. Through this paper, you will be able to build a clear selection logic, choose the most suitable epigenetic clock for your research topic, or plan the development path of a customized clock when necessary.

The Model Dilemma: Navigating the Proliferation of Epigenetic Clocks

Once upon a time, researchers only had to choose between a few epigenetic clocks. Today, faced with a dazzling array of models, decision-making without guidance is like groping in a maze. The key to solving this dilemma lies in understanding the intergenerational evolution history behind it. This is not a simple version upgrade, but a profound change in its core training objectives, construction methodology, and scientific connotation.

First-generation Clock: An Accurate Time and Age Predictor

Core mission and training goal: To predict the individual's time age (full age) with minimum error. They were born to answer a basic question: Can DNA methylation accurately reflect the passage of time?

Representative model:

• HorvathClock (2013): This is a landmark multi-organization clock, which can accurately predict the age in various tissues (such as blood, brain, and liver) and cell types, showing the universality of epigenetic aging.
• HannumClock (2013): Based on the development of blood samples, it shows extremely high accuracy in predicting blood age, and the average absolute error can be as low as 3-4 years.

Working principle: Using a machine learning algorithm (such as elastic network regression), hundreds of CpG loci with the strongest linear relationship with time and age were screened out from hundreds of thousands, and a powerful multiple regression prediction model was constructed.

Values and limitations:

• Value: They eloquently prove that epigenetic modification is a reliable yardstick to measure biological aging, which lays the foundation for the whole field. It is very suitable for evaluating tissue-specific aging and benchmarking a technology platform.
• Limitations: Their training goal is purely time and age, so they are better at capturing normal aging than abnormal or pathological aging related to disease and death risk. A clock that can accurately predict your time and age may not best reflect your healthy life.

Second-generation Clock: A Healthy Age Predictor Related to Disease Risk

Core Mission and Training Objective: Change the prediction target from pure time lapse to clinical phenotype, incidence, and all-cause mortality related to aging. They try to answer: How unhealthy are you?

Representative model:

• PhenoAge (2018): Its training target is not age, but a compound phenotypic age, an index that is composed of nine clinical blood biochemical indicators (such as creatinine, blood sugar, C-reactive protein, lymphocyte percentage, etc.) and can comprehensively reflect the degree of physiological system disorder.
• GrimAge (2019): This is another leap. It adopts an ingenious two-step method: firstly, it uses methylation data to predict the level of plasma proteins (such as growth differentiation factor -15GDF-15, lung surfactant protein DSFTPD) and smoking history (through DNAM-PACKYRS) related to mortality; Then, these predictors are combined into a final model. GrimAge is essentially a powerful proxy indicator of death risk.

Values and limitations:

• Value: The performance of the second-generation clock is usually significantly better than that of the first-generation clock in predicting all-cause mortality, cardiovascular disease, cancer, diabetes, and other health endpoints. They can effectively stratify risks and identify individuals with high biological age and high disease risk.
• Limitations: The model construction is more complicated, and its readings are complex, so it is necessary to understand the integrated biological pathways behind it (such as inflammation, metabolism, and tissue damage) when interpreting it.

Third-generation clock: A Dynamic Index to Quantify Speed of Aging

Core mission and training goal: Measure the rate of aging (Pace), not the accumulated age. They don't ask how old you are now. But how fast are you aging at present?

Representative model: DunedinPACE(2022), and some versions of GrimAge2 also incorporate similar concepts.

Working principle: Based on longitudinal research data (the same group of participants was followed up for many years, and DNA methylation was detected), the actual growth of their biological age per year was calculated by analyzing the dynamic change trajectory of their methylation patterns during the follow-up period. It is essentially a derivative of aging.

Values and limitations:

• Value: It is extremely sensitive to short-term changes and is the ideal endpoint of interventional clinical trials. It can detect the acceleration or deceleration of the aging pace caused by lifestyle changes, drug intervention, or major stress events within a few months to a year or two.
• Limitations: Its construction depends on high-quality and large-scale longitudinal queue resources, and its predictive value for ultra-long-term (such as ten years) healthy outcomes is still being accumulated and verified by the database.

Understanding the fundamental differences between these three models in their goals and missions is the logical starting point for making any wise choice.

Epigenetic disorder underlies epigenetic clock signals (Bertucci-Richter et al., 2024)

Aligning First-, Second- and Third-Generation Epigenetic Clocks with Research Objectives

Choosing a clock is not about pursuing the most advanced or complicated, but about finding the most suitable one. Specific research problems should be the only guide for model selection.

Estimate the Whole or Tissue-specific Aging Degree

Recommended choice: The first-generation clock (such as HorvathClock, HannumClock).

Reason: If you need to evaluate the global biological age of a rare patient, or directly compare the aging degree of a specific tissue (such as tumor and adjacent tissues, specific areas of the brain) relative to time and age, the first-generation clock is a direct and ideal choice. They provide a baseline reading of biological age, and the concept is intuitive.

Typical research situation:

• Biological age assessment of patients with premature senility syndrome (such as Hutchinson-Gilford syndrome).
• Comparison of brain aging between patients with Alzheimer's disease and healthy controls after death.
• To evaluate the difference in aging rate of different organs in natural aging or a disease state.

Predict Specific Health Risks

Recommended choice: the second-generation clock, especially GrimAge and its derivatives.

Reason: These models were designed from the beginning to capture the biological processes related to adverse health outcomes. If your research focuses on prognosis judgment, risk stratification, or exploring the erosion effect of some chronic exposure (such as long-term air pollution and poor diet) on healthy life, the second-generation clock provides the most direct and powerful evidence.

Typical research situation:

• In a large community cohort, baseline DNAmGrimAge was used to predict the risk of cardiovascular events in the next 10 years.
• To evaluate the accelerated epigenetic age of patients with metabolic diseases such as obesity and type 2 diabetes, and to correlate the risk of complications.
• To explore the relationship between GrimAge and the overall survival of patients in the tumor cohort.

Measure the Speed of Aging and Evaluate Short-term Intervention

Recommended choice: the third-generation clock, such as DunedinPACE.

Reason: In an intervention study or a short-term longitudinal study, the core is the dynamic change trend of participants' aging process, not the cumulative aging amount at a static time point. Pace clock is extremely sensitive to this change rate, and can detect the change of aging rate in a relatively short period (such as 6 months to 2 years).

Typical research situation:

• A randomized controlled trial of calorie restriction or high-intensity interval training for 12 months, the main endpoint is the change in DunedinPACE.
• To observe the short-term effects of major psychosocial stress (such as caregivers' long-term care stress) on the aging rate.
• In drug development, as an exploratory biomarker of phase I or phase II clinical trials, the anti-aging potential of candidate drugs is preliminarily verified.

Blood and myometrial DNA methylation using three epigenetic clocks (Erickson et al., 2022)

A Data-Driven Framework for Model Selection

After defining the macroscopic correspondence between the research goal and the clock generation, a more elaborate and data-driven operation framework can help you optimize your decision-making and even explore deeper biological insights.

Multi-clock Parallel Strategy and Consistency Analysis.

Core strategy: When the expected results are uncertain or exploratory research is conducted, it is strongly recommended to run multiple clocks of different generations on the same sample set at the same time (for example, simultaneously calculate HannumAge, PhenoAge, GrimAge, and DunedinPACE).

Value and interpretation:

• Enhance the robustness of the results: If multiple clocks from different generations (such as Hannum of the first generation and GrimAge of the second generation) all show consistent and significant age acceleration signals, then the research findings will be very stable and the conclusions will be highly reliable.
• Revealing biological heterogeneity: If different clocks give inconsistent or separated signals, it is a valuable discovery in itself. This phenomenon of signal separation can provide crucial clues and directions for the follow-up mechanism research.

Reference to Integrated Ideas

Forward thinking: Instead of betting on a single model, we should consider integrating the prediction results of multiple clocks to form a more robust aging index. In the study of animal models, a framework such as EnsembleAge has emerged, which integrates the prediction results of multiple basic clocks and forms a more stable and accurate composite clock by using an ensemble learning method.

For example, the standardized scores (Z-score) of GrimAge, PhenoAge, and DunedinPACE are weighted averages, or a principal component score is constructed to create a comprehensive healthy aging state index. This method can effectively suppress the inherent random error of a single model and comprehensively utilize the complementary biological information captured by different models, thus providing a more comprehensive and robust aging assessment.

Consideration of Technology Platform Compatibility

Key checkpoint: Before you finally decide to use a clock, be sure to confirm that it is fully compatible with the DNA methylation detection platform. Most of the published clocks are based on Illumina Human Methylation 450K or EPIC(850K) chips. If you use an updated platform (such as EPICv2) or a targeted sequencing technology (such as TIME-seq), some key CpG sites in the original model may not be covered.

Solution: You need to carefully check the site list or find the model version that has been successfully migrated to the new platform. In some cases, it may be necessary to map sites or use a filling algorithm, but this may introduce uncertainty.

Causal estimates from genetically predicted AVS to epigenetic age (Pan et al., 2024)

When Do You Need a Custom Epigenetic Clock

Select with confidence, or built for precision — For most human population studies based on blood samples, the existing and fully verified clocks (especially the second and third generations) are powerful and reliable enough. However, when your research touches the boundary of the existing model, building a customized clock will be sublimated from an option to a strategic necessity to promote scientific discovery.

Research Object is A Non-standard Species

• Specific situation: The research involves model animals, economic animals, or companion animals such as mice, rats, dogs, and nonhuman primates.
• The root cause: The specific CpG locus on which the human clock depends may not exist in the animal genome, with low sequence conservation, or its methylation pattern with age is completely different from that of humans. It is unreasonable in biology to directly apply the human clock to animal samples, and the result cannot be interpreted.
• Solution: For the specific species you study, systematically collect healthy samples of different ages covering the whole life cycle, and build a species-specific epigenetic clock from scratch. This is very important for objectively evaluating the efficacy of preclinical anti-aging drugs.

Research Focus is Non-blood Tissue

• Specific situation: It is necessary to directly and accurately evaluate the aging of specific tissues such as the brain, liver, kidney, skin, and fat.
• Root cause: Although Horvath's multi-organization clock is a groundbreaking universal tool, it is essentially a compromise of multi-organization signals. A clock specially trained in a specific tissue can more sensitively and specifically capture the unique aging pattern of the tissue and the changes in related gene regulatory pathways.
• Solution: Build a tissue-specific clock. This can greatly improve your ability to find biomarkers of age-related diseases (such as fibrosis in the liver and neurodegeneration in the brain) in this tissue.

Study Rare diseases or Special People

• Specific situation: We are concerned about a rare disease that is seriously under-represented in the universal clock training queue, or a specific ethnic group with a unique genetic background, environmental exposure, and lifestyle.
• Root cause: The training data of the universal clock may not cover the specific pathophysiological process of the disease or population, which leads to the model being insensitive or inaccurate in evaluating its aging state and unable to detect the real aging acceleration signal.
• Solution: Build a disease-specific or population-specific clock. Such a clock may reveal the unique aging trajectory driven by specific causes that cannot be captured by general clocks, and provide a new tool for understanding the disease mechanism and evaluating intervention measures.

Adopting the Emerging Technology Platform

• Specific situation: EPICv2 chip, more cost-effective targeted methylation sequencing (such as TIME-seq) or genome-wide bisulfite sequencing are used.
• The root cause: the probe design, coverage, and technical principle of the new platform may change, resulting in the loss of key sites or the change of data distribution in the old model, and the direct application will seriously affect the prediction accuracy.
• Solution: Using the high-quality data generated by the new platform, fine-tune and adapt the existing optimal model through advanced statistical methods such as transfer learning, or retrain an optimal clock directly based on the new platform data to ensure the scientific research and the reliability of the results.

Construction Process of A Customized Clock

If research really needs to be customized, a rigorous and professional construction process usually includes the following core links:

• Strategic research design and sample collection: This is the cornerstone of success. Careful planning is required to ensure that the sample can cover the entire age/pathological pedigree of the species, tissues, or disease states you are concerned about, accompanied by high-quality, standardized metadata.
• Standardized wet experiment and high-quality data generation: DNA extraction, bisulfite conversion, and Qualcomm methylation detection are carried out by strictly optimized standardized processes to ensure the reliability and repeatability of the original data.
• Robust model training and strict verification: an advanced machine learning process is used for feature (CpG locus) screening and model training, and the performance (such as accuracy, robustness, and predictive ability) of the clock must be verified in a completely independent sample set (test set) in multiple dimensions and strictly.
• In-depth biological interpretation and prospective application: In-depth analysis of the biological functions (such as genes and regulatory pathways) of the key CpG loci that constitute the clock, and application of your new clock to specific scientific problems to verify its practical value.

Summary-statistics-based Mendelian randomization (MR) analyses for the associations of different epigenetic clocks with bladder cancer risk (Deng et al., 2024)

Conclusion

In the increasingly rich toolbox of the epigenetic clock, there is no master key that can solve all problems. The successful selection stems from the profound insight into the intergenerational mission of the model, the accurate calibration with the core objectives of the research, and the foresight and courage to embrace the customization strategy in the frontier exploration.

The first-generation clock is the cornerstone of the field, which proves the feasibility of measuring time with methylation. The second generation clock deeply binds aging and health risks, which greatly enhances the clinical relevance of the study. The third-generation clock allows us to dynamically monitor the aging flow rate for the first time, which brings revolutionary tools for intervention research. By adopting a data-driven parallel strategy or integration idea, we can maximize the efficiency of existing model arrays and capture new scientific clues from the inconsistency between models. When our scientific exploration sails to the blank area of the existing map, whether it is a new species, a special organization, a unique disease group, or an innovative technology platform, building a customized clock will change from an option to an inevitable choice to lead the scientific research on precise aging.

FAQ

1. What's the core difference between first-, second-, and third-generation epigenetic clocks?

First-gen (e.g., Horvath) predicts chronological age; second-gen (e.g., GrimAge) links to disease/mortality; third-gen (e.g., DunedinPACE) measures aging speed.

2. Which clock fits risk prediction research (e.g., cardiovascular disease)?

Second-gen clocks like GrimAge—they're trained on health outcomes and excel at stratifying disease/mortality risk.

3. Can I use a human epigenetic clock for animal studies (e.g., mice)?

No. Human clocks rely on species-specific CpG sites; customize a species-specific clock for valid results.

4. When should I run multiple clocks instead of one?

For exploratory research or to boost result robustness—consistent signals across clocks confirm reliability; inconsistencies hint at biological heterogeneity.

5. Is a customized clock necessary for non-blood tissues (e.g., brain)?

Recommended. Tissue-specific clocks better capture unique aging patterns than multi-tissue clocks (e.g., Horvath) which compromise across tissues.

Related reading

• Epigenetic Clocks 101: Biological vs Chronological Age in Aging Research
• Designing Epigenetic Clock Studies: Endpoints, ΔAge and Power Considerations
• Choosing Epigenetic Clock Models: Horvath, Hannum, PhenoAge, GrimAge and Beyond
• Using Epigenetic Age Acceleration to Study Environmental and Lifestyle Exposures

References

1. Bertucci-Richter EM, Shealy EP, Parrott BB. "Epigenetic drift underlies epigenetic clock signals, but displays distinct responses to lifespan interventions, development, and cellular dedifferentiation." Aging (Albany NY). 2024 16(2): 1002-1020.
2. Erickson EN, Knight AK, Smith AK, Myatt L. "Advancing understanding of maternal age: correlating epigenetic clocks in blood and myometrium." Epigenetics Commun. 2022 2: 3.
3. Pan W, Huang Q, Zhou L, et al. "Epigenetic age acceleration and risk of aortic valve stenosis: a bidirectional Mendelian randomization study." Clin Epigenetics. 2024 16(1):41.
4. Deng Y, Tsai CW, Chang WS, et al. "The Significant Associations between Epigenetic Clocks and Bladder Cancer Risks." Cancers (Basel). 2024 16(13): 2357.