Aging Clocks Catalog · Understand the measurements
Clock types and what they measure
A clock’s input, training target and biological scope are three separate choices. Keeping them separate prevents misleading comparisons.
Name the input, target and scope separately
DNA methylation, clinical chemistry, proteins, RNA, metabolites, glycans, images, activity or another specified measurement.
Calendar-age resemblance, an age-scaled phenotype, mortality-related risk, longitudinal pace or reference-state distance.
Whole person, organ, tissue, cell type or species. Scope does not identify the measurement or target.
A methylation-based systems score and a circulating-protein organ score may share an organ label but not the same inputs or estimand.
A raw age gap is often predicted age minus chronological age; an age-acceleration residual instead depends on a fitted reference relationship. “Years” can also be a risk-equivalent scale. Do not compare these labels until their definitions match. See the worked numerical distinctions. [M04; SC6]
An organ label may refer to tissue-enriched circulating proteins, MRI anatomy, or a methylation proxy for another biological layer. Enrichment helps select features; it does not prove that a circulating protein came only from that organ or diagnoses damage there. Changing the training target can also change what two scores mean even when their input assay is the same. [V04; V05; V09; M21]
Selected anchors—not interchangeable models
| Anchor | Measurement and target | What the name does not settle |
|---|---|---|
| Horvath 2013 | 353-CpG DNA-methylation age estimator developed across many tissues/cell types. [M01] | A multi-tissue development set does not establish equal accuracy in every tissue, disease or life stage. Preserve the original transformations and normalization. |
| Hannum 2013 | Whole-blood DNAm age model; original study profiled 656 people aged 19–101. [M02] | The title’s “rates” language does not make it a model trained on repeated multisystem trajectories. Check calibration before changing tissue. |
| Clinical Phenotypic Age → DNAmPhenoAge | Clinical version uses age plus nine routine markers on a mortality-related age scale. DNAmPhenoAge uses 513 CpGs to approximate that phenotype. Clinical development/validation used NHANES; the DNAm stage used InCHIANTI and further cohorts. [M03] | These are different measurements and stages, not two spellings of one blood test. Clinical, DNAm and principal-component variants need distinct labels. |
| GrimAge2 | Blood DNAm mortality-oriented composite with proxies for proteins and smoking; version 2 adds log-CRP and log-HbA1c proxies. [M04] | Proxy outputs are not measured clinical concentrations. AgeAccelGrim2 is residualized; version 1, version 2 and principal-component variants cannot be pooled silently. |
| DunedinPACE | 173 selected CpGs approximate longitudinal pace based on 19 indicators measured at ages 26, 32, 38 and 45 in the Dunedin cohort. [M05] | Reference normalization uses more probes than the 173 predictors. The final feature count alone does not define a runnable input matrix. |
| KDM / homeostatic dysregulation | KDM uses reference biomarker–age relations and variance. Homeostatic dysregulation measures departure from a chosen reference, commonly with Mahalanobis distance. [T01] | KDM is a family of specifications; freeze the fit and biological-age variance. A distance measure is not automatically years of age. |
| ProtAge | 204-protein Olink-based LightGBM age model; UK Biobank development with external evaluation in China Kadoorie Biobank and FinnGen. [SC4] | The original model is gradient boosting, despite the trial paper’s deep-learning description. Plasma development does not establish serum or platform equivalence. |
| PAC / Goeminne OrganAge | PAC uses age and 128 proteins in a mortality model with Gompertz age mapping. Goeminne’s OrganAge family includes age- and mortality-oriented models. [SC6; SC7] | Similar age-like outputs need not share targets. PAC’s reported 70:30 split was within UK Biobank, not an independent external cohort. Oh’s organ models are a different implementation. |
The wider map
These families belong in the catalog, but their evidence is not equally mature. Abstract-level coverage supports identification of a method and its boundary—not a performance league table.
| Family / scope | Primary example | What the evidence stops short of |
|---|---|---|
| Bulk transcriptomics | Peters et al.: age-associated blood gene expression across 14,983 people and transcriptomic age measures. [M07] | Expression and cell composition depend on specimen/state. Retain mechanistic and specimen-matched uses; no verified locked clinical add-on comparison was extracted to justify acquiring expression data solely for risk selection. [V14] |
| Model-organism transcriptomics | BiT age uses binarized expression, prominently in C. elegans; lifespan metadata help rescale worm targets across conditions. [M08] | Target construction matters for intervention interpretation. This is not human treatment-induced lifespan evidence. |
| Single-cell / single-nucleus | Human brain cell-type clocks: 73,941 nuclei from 31 postmortem prefrontal-cortex donors. [M09] | 31 donors are not 73,941 independent people. Donor-level validation and living-person transfer remain separate questions. |
| Metabolomics / lipidomics | MetaboAge estimates age from NMR profiles; Deelen’s 14-biomarker NMR score predicts mortality risk. [M10; M11] | Age and mortality are different targets; routine lipids do not constitute the full NMR model. Selected NMR and LC–MS results justify research comparisons, but validation overlap, cohort/timepoint scaling and routine alternatives affect the claim. See the four-model clinical comparison. [V01; V03; V07] |
| Glycomics | IgG glycan-age research in several European populations links profiles with age and metabolic traits. [M12] | Global IgG profiles vary with age and geography. Separate plasma N-glycan risk research reports external diabetes evaluation in FINRISK, but full comparator/gain/calibration details were unavailable. Total-plasma N-glycans, IgG glycans and GlycA are not interchangeable; that risk result does not validate a GlycanAge service. [V11; V12] |
| Physiological / functional / digital | Pyrkov et al.: week-long NHANES accelerometry, age prediction and survival-oriented modeling. [M13] | Function can be a direct endpoint rather than a clock. PpgAge adds device-specific Apple-study wrist waveforms and pretrained representations—not any wearable’s pulse summary. No independent-device actionable risk-model validation was established; trained weights were not supplied by the inspected access material. [V10] |
| Imaging | Retinal-image age gap had an adjusted all-cause mortality association, but not significant cardiovascular- or cancer-mortality associations. [M14] | No retinal diagnosis or screening benefit follows. Brain and facial image ages need their own evidence, not borrowed retinal validation. |
| MRI with a longitudinally trained target | DunedinPACNI learns a multisystem pace-related phenotype from age-45 structural MRI in 860 Dunedin participants, with association tests in ADNI, UK Biobank and BrainLat. Comparators include brainageR gaps and hippocampal/ventricular volumes. [V09] | A relative quantity without clear units, not age-years or a personal risk probability. Useful research access for existing suitable MRI does not establish a reason to order a new aging scan. A larger combined hazard contrast is not an AUC or calibration gain. |
| Organ-proteomic prognosis | Oh and Wang provide newer plasma-protein transfer evidence; Whitehall II provides external long-term disease associations. [V04; V05; V06] | Externality has layers: score transport and outcome association do not establish a locked combined clinical-risk model’s external performance. Whitehall’s abstract reports mostly cross-organ associations, limiting literal organ-diagnosis interpretations. |
| Microbiome | Taxonomic age profiles (Galkin); species/pathway multi-view models with explicit geographic analyses (Chen). [M15; M16] | Myers et al.’s matched shotgun/16S comparison reports age MAE 8.68 versus 9.85 years. That is birthday prediction, not added clinical-outcome value. Cohort/geographic structure and independent-person handling of repeat samples remain essential; more sequencing is not itself a clinical aging-test justification. [V13] |
| Other epigenetic / chromatin | sc-ChromAging (9 May 2026): single-cell chromatin-accessibility clocks in a Chinese cohort. [M19] | A different regulatory layer from DNAm; abstract-level emerging-method evidence, not a full clinical audit. |
| Replication / mitotic history | epiTOC aggregates methylation at selected Polycomb-associated sites as a mitotic-history-related score. [M17] | Cell division history is not remaining lifespan or a universal person-level aging measure. |
| Cross-species | Mammalian DNAm clocks across 185 species include chronological and relative-age formulations. [M18] | Lifespan, gestation and maturity metadata in targets matter; conserved prediction does not imply interchangeable treatment effects. |
| Multimodal / organ / systems | Oh’s circulating-protein organ models, DNAm-based Systems Age and OMICmAge illustrate different input/target arrangements. [M20; M21; M22] | OMICmAge deployment uses DNAm-derived inputs—not direct measurement of the customer’s whole proteome and metabolome. More outputs do not guarantee independent information. |
Use the map to select a comparison, not a winner. Compatible archived proteomics or MRI can make an extra score a low-acquisition-burden research option. New collection adds assay, processing, failure and interpretation burdens. The addition-versus-replacement guide and worked risk examples show which evidence would justify each choice.
Not every aging-related number is a clock
Telomere length, an inflammatory protein, grip strength, frailty and a wellness questionnaire can be useful biomarkers or clinical measures without being trained age estimators. A DNAm proxy for a protein or telomere-related quantity is not its direct assay. This catalog classifies by the actual method and target, not the phrase “health age.” Conventional measures can be the comparator that a more expensive clock must improve upon.
Ask “better at what?” before “which is best?” The 2025 benchmark abstract reports little relationship between chronological-age accuracy and mortality-prediction capacity across 39 biomarkers in more than 20,000 participants. Full methods were unavailable; no detailed ranking is adopted. [M06]
Sources and reading limits
Source labels distinguish primary research, seller documents and implementation notes. The access descriptions below refer to the evidence review dated above, not a new replication.
M01. Horvath (2013), DNA methylation age of human tissues and cell types Primary abstract and relevant model/tissue passages inspected; not every supplement or later variant.
M02. Hannum et al. (2013; online 2012), Genome-wide methylation profiles Primary abstract and figure descriptions inspected; no full coefficient audit.
M03. Levine et al. (2018), An epigenetic biomarker of aging for lifespan and healthspan Clinical selection, units/coefficient table, methylation stage and validation passages inspected; no raw-data reanalysis.
M04. Lu et al. (2022), DNA methylation GrimAge version 2 Primary version, training and multi-cohort validation passages inspected; no commercial-version equivalence inferred.
M05. Belsky et al. (2022), DunedinPACE Primary longitudinal target, normalization, technical/cross-platform reliability and relevant validation text inspected; model not executed.
M06. Ying et al. (2025), A unified framework for systematic curation and evaluation of aging biomarkers Author-institution abstract/metadata only; detailed methods and rankings not adopted.
M07. Peters et al. (2015), The transcriptional landscape of age in human peripheral blood Primary abstract and relevant expression/prediction discussion inspected; no full pipeline audit.
M08. Meyer and Schumacher (2021), BiT age Primary binarization, worm target-rescaling and validation passages inspected; no human clinical-utility inference.
M09. Muralidharan et al. (2025), Human Brain Cell-Type-Specific Aging Clocks Primary abstract/opening results and availability text; complete donor-split and per-cell methods not audited.
M10. van den Akker et al. (2020), Metabolic age based on the BBMRI-NL NMR repository Author-laboratory abstract/metadata only; used to distinguish targets, not rank performance.
M11. Deelen et al. (2019), A metabolic profile of all-cause mortality risk Primary results, conventional comparator, FINRISK evaluation and scaling limitation inspected; reported models not rerun.
M12. Krištić et al. (2014; online 2013), Glycans are a novel biomarker of chronological and biological ages Primary development, external-population and small longitudinal-subset results; not validation of a pinned 2026 product.
M13. Pyrkov et al. (2018), Extracting biological age from biomedical data Primary abstract and activity/model-development passages; no device-service certification.
M14. Zhu et al. (2022), Retinal age gap as a predictive biomarker for mortality risk Complete primary abstract, including outcome-specific nulls; full methods and deployment validation not inspected.
M15. Galkin et al. (2020), Human Gut Microbiome Aging Clock Primary abstract only; no independent microbiome processing or clinical-utility claim.
M16. Chen et al. (2022), Human gut microbiome aging clocks through multi-view learning Primary abstract and relevant geography, input and validation sections; code/data not run.
M17. Yang et al. (2016), Correlation of an epigenetic mitotic clock with cancer risk Primary score definition and development/validation passages; no conversion to person-level age.
M18. Lu et al. (2023), Universal DNA methylation age across mammalian tissues Primary abstract, target/scope and limitation passages; no treatment-equivalence claim across species.
M19. Wei et al. (2026), sc-ChromAging Published 9 May 2026. Primary abstract/metadata only; emerging-method coverage, not a full model audit.
M20. Oh et al. (2023), Organ aging signatures in the plasma proteome Primary construction, tissue-enrichment, platform and cognitive-progression comparator passages; no decision-impact trial identified in that material.
M21. Chen et al. (2026), OMICmAge Published 25 February 2026. Primary development, external validation, replicate, limitation, access and conflict sections; sponsored/company-affiliated work with patent interests.
M22. Sehgal et al. (2025), Systems Age Version of record: 15 September 2025. Abstract/metadata and primary update linkage inspected; detailed rankings and current product equivalence withheld.
T01. BioAge R package README and relevant source files inspected, not executed. DESCRIPTION 0.1.0 declares GPL-3; package paper was identified, not independently read.
SC4. Argentieri et al. (2024), Proteomic aging clock predicts mortality and disease risk Primary training, external validation and covariate passages inspected through publisher-provided text delivered on ResearchGate. No model execution.
SC6. Kuo et al. (2024), Proteomic aging clock (PAC) Primary cohort, selection, Gompertz target and train/test passages; full coefficients and ancillary-trial artifact not verified.
SC7. Goeminne et al. (online 2024; issue 2025), Plasma protein-based organ-specific aging and mortality models Primary abstract and publisher results excerpts, not full coefficient tables or frozen trial implementations.
Added V/I evidence and access observations: . Published findings, code inspection and catalog arithmetic are separate; no new clinical or molecular model was rerun.
V04. Oh et al. (2025), Organ-proteomic development and transfer study Relevant full UK Biobank/Stanford, paired-platform and repeat-visit methods/results inspected; no supplementary-coefficient or calibrated-risk replication. Commercial/patent interests disclosed; external cohort evidence is not independent service certification.
V05. Wang et al. (online 2025; 2026 issue), Organ-specific proteomic aging clocks Relevant full cohort, assay, development and Figure 6 comparison passages inspected. No combined-model rerun, calibration audit or recovery of all exclusions. Authors declare no competing interests.
V06. Kivimäki et al. (2025), Whitehall II organ-age follow-up Primary abstract and author list inspected; full methods unavailable. Reported design and cross-organ patterns only, not detailed incremental discrimination or calibration. Investigators overlap the originating work.
V09. Whitman et al. (2025), DunedinPACNI Relevant development, comparator, outcome, limitations and availability passages in the author-hosted publisher PDF inspected; Extended Data Figure 6/caption checked visually. No raw imaging, supplementary numerical table or model execution. Development-team evidence, not consumer validation.
V01. van Holstein et al. (2024), Metabolomic mortality scores in older patients with solid tumors Relevant full methods/results and Table 4 (publisher PDF page 9) inspected. Four combined-model summaries audited; no participant-level ROC, calibration or clinical-benefit reproduction. Several score developers are coauthors.
V03. Deelen et al. (2019), A metabolic profile of all-cause mortality risk Original prediction/evaluation and discovery-overlap sensitivity passages rechecked through primary article text delivered on ResearchGate; unrelated citing material excluded. Estonian evaluation-score fitting and FINRISK evaluation distinguished. No score rerun.
V07. Lau et al. (2026), A metabolomic clock of population ageing Published 21 July. Relevant full training/cohort, LC–MS assay, residualization and mortality-comparison passages inspected. No controlled Airwave/TILDA data acquired; no model or raw ROC reproduction.
V10. PpgAge study (2025) Relevant full device, age/health sample, model, outcome and availability passages inspected. Apple study/developer context; linked repository README was not readable through the attempted route. Untrained code and synthetic material are not trained weights or participant data.
V11. Štambuk et al. (2020), Global IgG glycosylation patterns Relevant primary geographic/age, specimen and methods passages inspected. Genos affiliations/commercial interests; no universal calibration factor inferred.
V12. Wittenbecher et al. (2020), Plasma N-glycans and cardiometabolic risk Complete primary abstract inspected. Full comparator definitions, external add-on magnitudes and calibration details unavailable; no numerical gain or clinical threshold inferred.
V13. Myers et al. (2025), Microbiome age-prediction comparison Relevant full paired 16S/shotgun results, cohort structure and repeated-sample splitting inspected. No microbiome matrix, training or prospective clinical-risk validation rerun.
V14. Peters et al. (2015), Blood transcriptomic aging Primary abstract record inspected; earlier expression/prediction passages are inherited in M07. No new locked clinical add-on performance result extracted.
Next: Turn the map into a study choice or learn how to compare outputs.