Uncovering dynamics of age-related epigenetic changes with an interpretable deep-learning framework

Uncovering dynamics of age-related epigenetic changes with an interpretable deep-learning framework | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Uncovering dynamics of age-related epigenetic changes with an interpretable deep-learning framework Aaron Lin, Ilinca Giosan, Andrea Aparicio, Tao Guo, Laura Balagué-Dobón, and 9 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7614352/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 13 Mar, 2026 Read the published version in npj Aging → Version 1 posted 19 You are reading this latest preprint version Abstract Aging is the strongest risk factor for chronic diseases such as cardiovascular disease, Alzheimer’s, and cancer. DNA methylation (DNAm) clocks offer a promising measure of biological age, but most rely on linear models that miss non-linear dynamics and CpG interactions. To address this, we developed a deep neural network (DNN)-based DNAm clock trained on 29,167 samples profiled on Illumina EPIC v1.0 and v2.0 arrays. Using 12,234 CpGs selected through sex-and age-stratified correlations, our model achieved high accuracy (1.89 years) and outperformed published deep learning and elastic net based epigenetic clocks in a separate validation cohort. Using Shapley Additive Explanations (SHAP), we further uncovered phase-structured, wave-like dynamics in age-influential CpGs: an early-life module, a midlife transition, and late-life remodeling, with distinct timings by sex. These epigenetic waves cohere with non-linear, multi-omic “aging waves” reported in proteomics and longitudinal omics. SHAP further enabled interpretable CpG attribution, revealing structured, sex-specific aging phases: early-life male clocks involved developmental pathways, while female clocks emphasized cytoskeletal regulation; late-life divergence included immune activation in males and transcriptional remodeling in females. Our framework thus unites accuracy with mechanistic interpretability, revealing sex-specific windows when molecular aging reconfigures most rapidly. Health sciences/Biomarkers Biological sciences/Computational biology and bioinformatics Biological sciences/Genetics Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Full Text Additional Declarations Competing interest reported. A.L., I.G., L.B.D., N.C.G, S.H., K.S., R.S., and V.B.D. are all employees of TruDiagnostic Inc. L.P.C. is an employee for Shift Bioscience and serves as a consultant to for profit organization TruDiagnostic Inc. J.L.S. is a scientific advisor to for profit organizations TruDiagnostic Inc, Precision Inc. and Ahara Inc. No other competing interests are noted among the authors. Supplementary Files FigS1.tiff FigS2.tiff FigS3.tiff Cite Share Download PDF Status: Published Journal Publication published 13 Mar, 2026 Read the published version in npj Aging → Version 1 posted Editorial decision: Revision requested 06 Nov, 2025 Reviews received at journal 04 Nov, 2025 Reviews received at journal 03 Nov, 2025 Reviews received at journal 03 Nov, 2025 Reviews received at journal 02 Nov, 2025 Reviews received at journal 31 Oct, 2025 Reviewers agreed at journal 29 Oct, 2025 Reviewers agreed at journal 29 Oct, 2025 Reviews received at journal 28 Oct, 2025 Reviewers agreed at journal 27 Oct, 2025 Reviewers agreed at journal 24 Oct, 2025 Reviewers agreed at journal 24 Oct, 2025 Reviewers agreed at journal 23 Oct, 2025 Reviewers agreed at journal 23 Oct, 2025 Reviewers agreed at journal 23 Oct, 2025 Reviewers invited by journal 23 Oct, 2025 Editor assigned by journal 17 Sep, 2025 Submission checks completed at journal 15 Sep, 2025 First submitted to journal 14 Sep, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7614352","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":516645807,"identity":"7bc76e0a-a4af-4868-809f-702b42307a5b","order_by":0,"name":"Aaron Lin","email":"","orcid":"","institution":"TruDiagnostic","correspondingAuthor":false,"prefix":"","firstName":"Aaron","middleName":"","lastName":"Lin","suffix":""},{"id":516645808,"identity":"01e3529e-a1d0-478a-80e1-e684f167ad11","order_by":1,"name":"Ilinca Giosan","email":"","orcid":"","institution":"TruDiagnostic","correspondingAuthor":false,"prefix":"","firstName":"Ilinca","middleName":"","lastName":"Giosan","suffix":""},{"id":516645809,"identity":"a13dcd6e-c008-4636-8632-23678426c45d","order_by":2,"name":"Andrea Aparicio","email":"","orcid":"","institution":"Harvard Medical School","correspondingAuthor":false,"prefix":"","firstName":"Andrea","middleName":"","lastName":"Aparicio","suffix":""},{"id":516645810,"identity":"b69e1a49-6901-43c0-b39d-1717c94fe0e7","order_by":3,"name":"Tao Guo","email":"","orcid":"","institution":"Harvard Medical School","correspondingAuthor":false,"prefix":"","firstName":"Tao","middleName":"","lastName":"Guo","suffix":""},{"id":516645811,"identity":"f2fa918d-99a3-4173-807c-71acee9f0aa2","order_by":4,"name":"Laura Balagué-Dobón","email":"","orcid":"","institution":"TruDiagnostic","correspondingAuthor":false,"prefix":"","firstName":"Laura","middleName":"","lastName":"Balagué-Dobón","suffix":""},{"id":516645812,"identity":"d51d6d29-467c-4983-a6eb-42ef912c48de","order_by":5,"name":"Natàlia Carreras-Gallo","email":"","orcid":"","institution":"TruDiagnostic","correspondingAuthor":false,"prefix":"","firstName":"Natàlia","middleName":"","lastName":"Carreras-Gallo","suffix":""},{"id":516645813,"identity":"c984238e-6f3c-48a9-84cc-fed4d014593d","order_by":6,"name":"Sayf Hassouneh","email":"","orcid":"","institution":"TruDiagnostic","correspondingAuthor":false,"prefix":"","firstName":"Sayf","middleName":"","lastName":"Hassouneh","suffix":""},{"id":516645814,"identity":"eadc4f01-f29e-466e-a97d-e2f152aa8274","order_by":7,"name":"Kirsten Seale","email":"","orcid":"","institution":"TruDiagnostic","correspondingAuthor":false,"prefix":"","firstName":"Kirsten","middleName":"","lastName":"Seale","suffix":""},{"id":516645815,"identity":"e74c6a45-2c06-4b67-9a23-42ce65945c3e","order_by":8,"name":"Alex Kowalewski","email":"","orcid":"","institution":"TruDiagnostic","correspondingAuthor":false,"prefix":"","firstName":"Alex","middleName":"","lastName":"Kowalewski","suffix":""},{"id":516645816,"identity":"9ab10b7c-73b8-4e9c-887a-01133f483cb5","order_by":9,"name":"Brent Harrison","email":"","orcid":"","institution":"University of Kentucky","correspondingAuthor":false,"prefix":"","firstName":"Brent","middleName":"","lastName":"Harrison","suffix":""},{"id":516645817,"identity":"86a97d80-8902-4c61-bc82-856eef5d0e66","order_by":10,"name":"Ryan Smith","email":"","orcid":"","institution":"TruDiagnostic","correspondingAuthor":false,"prefix":"","firstName":"Ryan","middleName":"","lastName":"Smith","suffix":""},{"id":516645818,"identity":"b4a56a35-5a60-4054-ba04-31797c5c0456","order_by":11,"name":"Jessica Lasky-Su","email":"","orcid":"","institution":"Harvard Medical School","correspondingAuthor":false,"prefix":"","firstName":"Jessica","middleName":"","lastName":"Lasky-Su","suffix":""},{"id":516645819,"identity":"f1b5ab9e-0788-4176-97c3-766b168641e0","order_by":12,"name":"Lucas Paulo Lima Camillo","email":"","orcid":"","institution":"Shift Bioscience","correspondingAuthor":false,"prefix":"","firstName":"Lucas","middleName":"Paulo Lima","lastName":"Camillo","suffix":""},{"id":516645820,"identity":"c69eae3b-03e7-4046-91ef-d0f59b55231f","order_by":13,"name":"Varun B. Dwaraka","email":"data:image/png;base64,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","orcid":"","institution":"TruDiagnostic","correspondingAuthor":true,"prefix":"","firstName":"Varun","middleName":"B.","lastName":"Dwaraka","suffix":""}],"badges":[],"createdAt":"2025-09-14 19:08:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7614352/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7614352/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41514-026-00358-w","type":"published","date":"2026-03-13T15:59:38+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":91577874,"identity":"5ae480ef-b951-447d-bdda-d8ab9486d2fd","added_by":"auto","created_at":"2025-09-18 02:55:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":210893,"visible":true,"origin":"","legend":"\u003cp\u003eFeature selection and model training strategy. (A) Number of CpGs exceeding a given Spearman correlation threshold (r) with chronological age, stratified by age group (10–40, 40–50, 50–70, 70–100 years). Older age groups show a larger number of CpGs with stronger age correlations, indicating that age-associated methylation signals intensify in later life. (B) UpSet plot showing intersections of the top 1% most age-correlated CpGs across age groups. While many CpGs are unique to individual age bins, a subset is shared across multiple life stages, reflecting both conserved and age-specific methylation signatures. (C) Schematic of the leave-one-cohort-out (LOCO) cross-validation approach used to assess model generalizability. In each of three models, one external dataset is excluded from training and used solely for testing, while the remaining datasets serve as the training set.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/1900548bc644483fa9561baa.png"},{"id":91577345,"identity":"544a1a44-65ee-466f-a020-dc7959323c7e","added_by":"auto","created_at":"2025-09-18 02:47:15","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":526243,"visible":true,"origin":"","legend":"\u003cp\u003eModel architecture, validation performance, and interpretability of the epigenetic aging predictor. (A) Overview of the modeling framework. A neural network was trained on DNA methylation profiles and sex to predict chronological age. SHAP (SHapley Additive exPlanations) values were used to identify CpGs most influential to predictions, which were then annotated via Gene Ontology (GO) enrichment to reveal biological pathways linked to epigenetic aging. (B) Predicted vs. actual age across four independent validation cohorts (BoA, CIBMTR, Internal, MGH), showing strong linear agreement and minimal deviation from the identity line. (C) Leave-one-cohort-out (LOCO) cross-validation mean absolute error (MAE) for MGH, BoA, and CIBMTR datasets, demonstrating generalizability across populations. (D) External validation results showing MAE and coefficient of determination (r²) for each cohort. Performance ranged from 1.19 to 3.02 years MAE with high r² (0.8559–0.9842). (E) Comparison of predicted vs. actual age across multiple aging clocks. Each scatter plot shows decimal chronological age (x-axis) vs. predicted age (y-axis), with MAE, RMSE, and Spearman ρ values annotated. Unique colors represent different clocks (legend in final panel); better performance is indicated by lower MAE/RMSE and higher ρ.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/615936e92b375d5d81c968cd.png"},{"id":91577347,"identity":"1f046d6a-6a57-4cb7-a6da-1b6abf0f4b41","added_by":"auto","created_at":"2025-09-18 02:47:15","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":250416,"visible":true,"origin":"","legend":"\u003cp\u003eHierarchical clustering of age groups by SHAP molecular signatures with bootstrap validation. Age groups (5-year bins; \u0026lt;20 and 85+ as open-ended) were clustered using Euclidean distance on SHAP-derived CpG influence profiles and Ward.D2 linkage. Multiscale bootstrap resampling (via pvclust) provides approximately unbiased (AU, red) and bootstrap probabilities (BP, green) above each dendrogram edge; AU ≥ 95% indicates strong support. Colored bars beneath the leaves reflect the four-phase schema used for interpretability (legend at right). The red dashed line in (A) marks the maximum clustering height used to define phases and is carried over at the same height in (B) and (C) to facilitate direct comparison of age-phase structure across sexes. (A) All individuals. Clustering resolves four distinct aging phases—early life (\u0026lt;35), early-midlife (35–44), late-midlife (45–64), and late life (≥65)—with AU-supported subclusters (e.g., \u0026lt;30, 35–44, 65–84). (B) Males. Despite the four-phase visual overlay, bootstrap support identifies two dominant clusters (\u0026lt;50 and ≥50), indicating a more compressed and binary age-phase structure in males. (C) Females. Bootstrap-supported splits yield three distinct molecular phases: early life (\u0026lt;40), midlife (40–64), and late life (≥65), with finer-grained AU-supported subclusters (e.g., \u0026lt;30, ≥75), reflecting more gradual and articulated age transitions.\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/dd2efcf3b3816a34d8c56a3c.png"},{"id":91577349,"identity":"debbcb12-efa4-4f54-8c31-7b60fc8792c4","added_by":"auto","created_at":"2025-09-18 02:47:15","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":365549,"visible":true,"origin":"","legend":"\u003cp\u003eFemale biological processes, molecular function, and cellular components associated with age-influential CpGs in early life (0–34 years) GO enrichment analysis was performed on the top 1000 CpGs most strongly correlated with age (high age-influence) and the bottom 1000 CpGs least correlated with age (low age-influence) in females during early life. Results are grouped by Biological Process (BP), Molecular Function (MF), and Cellular Component (CC). (A) BP enrichment of top CpGs revealed strong enrichment for actin filament organization, supramolecular fiber formation, and cytoskeletal remodeling pathways, indicating dynamic structural reorganization in early life. (B) BP enrichment of bottom CpGs showed enrichment for immune and hormonal regulatory processes, including thyroid hormone response, lipid metabolism, and epigenetic programming, suggesting these pathways exhibit more stable methylation during early life. (C) MF enrichment of top CpGs highlighted protein binding, ion channel binding, and voltage-gated channel regulation, while (D) broader MF enrichment included DNA-binding transcription factor activity and chromatin regulation, reflecting active developmental gene regulation. (E) CC enrichment of top CpGs mapped to intracellular anatomical structures, including the cytoskeleton, vesicles, and chromatin granules. (F) Bottom CpGs were enriched for membrane-associated compartments, such as the plasma membrane, synapse, and photoreceptor segments, suggesting relative stability in genes involved in structural signaling.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/37d052132b6d573f9325727a.png"},{"id":91577348,"identity":"4c83c771-3b08-458b-af8e-5247943afb24","added_by":"auto","created_at":"2025-09-18 02:47:15","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":448923,"visible":true,"origin":"","legend":"\u003cp\u003eMale biological processes, molecular function, and cellular components associated with age-influential CpGs in early life (0–49 years). GO enrichment analysis was performed on the top 1000 CpGs most strongly correlated with age (high age-influence) and the bottom 1000 CpGs least correlated with age (low age-influence) in males during early life. Terms are grouped by Biological Process (BP), Molecular Function (MF), and Cellular Component (CC). (A) BP enrichment of top CpGs showed strong enrichment for stress fiber assembly, actin filament organization, cytoskeletal remodeling, and cell junction organization, highlighting structural and mechanical pathways as key targets of early male epigenetic aging. (B) BP enrichment of bottom CpGs revealed genomic imprinting, chromatin organization, purine metabolism, and cell proliferation pathways, suggesting these processes maintain greater methylation stability in early life. (C) MF enrichment of top CpGs highlighted protein binding, ion channel regulation, and hydrolase activity, while (D) broader MF enrichment included DNA-binding transcription factor activity, RNA polymerase II-specific regulation, and chromatin interaction, consistent with dynamic transcriptional regulation. (E) CC enrichment of top CpGs mapped to intracellular and membrane-bound organelles, including vesicles, endosomes, and mitochondria. (F) Bottom CpGs were enriched for nuclear and chromatin-associated compartments, such as the nucleoplasm, replication fork, and ribonucleoprotein complexes, indicating relative stability in genes critical for genome maintenance.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/9af11d861a3cf682259034ab.png"},{"id":91577359,"identity":"aa70b6e4-075a-4aa1-8ca0-5f1b1b2a6956","added_by":"auto","created_at":"2025-09-18 02:47:15","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":469655,"visible":true,"origin":"","legend":"\u003cp\u003eFemale biological processes, molecular function, and cellular components associated with age-influential CpGs in late life (65-85 years). GO enrichment analysis was performed on the top 1000 CpGs most strongly correlated with age (high age-influence) and the bottom 1000 CpGs least correlated with age (low age-influence) in females during late life. Results are grouped by Biological Process (BP), Molecular Function (MF), and Cellular Component (CC). (A) BP enrichment of top CpGs revealed epigenetic and developmental pathways, including genomic imprinting, embryonic organ development, hippocampus development, and epigenetic programming of gene expression. Additional enrichment for immune and inflammatory signaling (e.g., type I interferon production, lipopolysaccharide response) indicates heightened immune regulation in late life. (B) BP enrichment of bottom CpGs showed cytoskeletal and cellular organization pathways, including actin filament bundle assembly, supramolecular fiber organization, and protein localization, suggesting stable methylation patterns in structural genes. (C) MF enrichment of top CpGs highlighted transcriptional regulation, particularly DNA-binding transcription factor activity and RNA polymerase II-specific sequence binding. (D) Broader MF enrichment of bottom CpGs also included DNA binding, chromatin regulation, and transcriptional activity, indicating these processes remain comparatively stable. (E) CC enrichment of top CpGs mapped to nuclear and chromatin-associated structures, such as the nucleoplasm, nuclear bodies, and protein–DNA complexes. (F) Bottom CpGs were enriched for intracellular and membrane-associated compartments, including the endoplasmic reticulum lumen, synapses, and photoreceptor segments.\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/bf93a9f2444fc0eb16c024f4.png"},{"id":91577364,"identity":"ec099768-67ba-477c-8bce-5285f0001b11","added_by":"auto","created_at":"2025-09-18 02:47:15","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":481240,"visible":true,"origin":"","legend":"\u003cp\u003eMale biological processes, molecular function, and cellular components associated with age-influential CpGs in late life (50-85 years) GO enrichment analysis was performed on the top 1000 CpGs most strongly correlated with age (high age-influence) and the bottom 1000 CpGs least correlated with age (low age-influence) in males during late life. Results are grouped by Biological Process (BP), Molecular Function (MF), and Cellular Component (CC). (A) BP enrichment of top CpGs revealed immune and inflammatory pathways, including regulation of heterotypic cell–cell adhesion, response to thyroid hormone, immunoglobulin-mediated signaling, and cytokine production (e.g., IL-1 and interferon-?), highlighting immune regulation as a key target of late-life male epigenetic aging. (B) BP enrichment of bottom CpGs showed cytoskeletal and proliferative processes, including actin filament and supramolecular fiber organization, cell adhesion, and cell cycle regulation, suggesting that these structural pathways remain more epigenetically stable in late life. (C) MF enrichment of top CpGs highlighted protein binding, JUN kinase activity, histone modification (H3K4/H3K27 methyltransferase activity), and chromatin regulation, pointing to dynamic transcriptional and intracellular signaling changes. (D) MF enrichment of bottom CpGs included hydrolase activity, lipid binding, transporter activity, and promoter-specific chromatin binding, indicating relative preservation of these structural and metabolic functions. (E) CC enrichment of top CpGs mapped primarily to nuclear and intracellular compartments, including the endoplasmic reticulum, vesicles, mitochondria, and blood microparticles. (F) Bottom CpGs were enriched for membrane-bound organelles and extracellular vesicles, including synaptic vesicles, ribonucleoprotein complexes, and host cell components, suggesting selective stability in structural and secretory domains.\u003c/p\u003e","description":"","filename":"Fig7.png","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/0b31698afc944f2912ccf4d1.png"},{"id":91578146,"identity":"b6d2a0ed-8a4a-43e5-b0f6-1c4ffe9b6eea","added_by":"auto","created_at":"2025-09-18 03:03:15","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":536874,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version.\u003c/p\u003e","description":"","filename":"Fig8.png","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/b988913f5483cae25fa1eace.png"},{"id":104740950,"identity":"7200a451-755e-4ec1-afd2-6c00e715571f","added_by":"auto","created_at":"2026-03-16 16:19:51","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1222382,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1_covered_74b7157a-c763-4d99-95bc-616016690b48.pdf"},{"id":91577875,"identity":"b948fa6e-a786-408d-bd86-542a209a968b","added_by":"auto","created_at":"2025-09-18 02:55:15","extension":"tiff","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2016638,"visible":true,"origin":"","legend":"","description":"","filename":"FigS1.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/41da9a2ee6e17432270858dd.tiff"},{"id":91577356,"identity":"d6477491-9c9c-4f94-8c7f-d055e71eee96","added_by":"auto","created_at":"2025-09-18 02:47:15","extension":"tiff","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":2016638,"visible":true,"origin":"","legend":"","description":"","filename":"FigS2.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/af4fb2d5fde85bb040686a03.tiff"},{"id":91577361,"identity":"22374fdb-92f5-4b6f-b71a-a19bcacc14f7","added_by":"auto","created_at":"2025-09-18 02:47:15","extension":"tiff","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":2016638,"visible":true,"origin":"","legend":"","description":"","filename":"FigS3.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7614352/v1/4614739b5c07f4009d3dd28d.tiff"}],"financialInterests":"Competing interest reported. A.L., I.G., L.B.D., N.C.G, S.H., K.S., R.S., and V.B.D. are all employees of TruDiagnostic Inc. L.P.C. is an employee for Shift Bioscience and serves as a consultant to for profit organization TruDiagnostic Inc. J.L.S. is a scientific advisor to for profit organizations TruDiagnostic Inc, Precision Inc. and Ahara Inc. No other competing interests are noted among the authors.","formattedTitle":"Uncovering dynamics of age-related epigenetic changes with an interpretable deep-learning framework","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"npj-aging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [npj Aging](https://www.nature.com/npjamd/)","snPcode":"41514","submissionUrl":"https://submission.springernature.com/new-submission/41514/3","title":"npj Aging","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7614352/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7614352/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Aging is the strongest risk factor for chronic diseases such as cardiovascular disease, Alzheimer’s, and cancer. DNA methylation (DNAm) clocks offer a promising measure of biological age, but most rely on linear models that miss non-linear dynamics and CpG interactions. To address this, we developed a deep neural network (DNN)-based DNAm clock trained on 29,167 samples profiled on Illumina EPIC v1.0 and v2.0 arrays. Using 12,234 CpGs selected through sex-and age-stratified correlations, our model achieved high accuracy (1.89 years) and outperformed published deep learning and elastic net based epigenetic clocks in a separate validation cohort. Using Shapley Additive Explanations (SHAP), we further uncovered phase-structured, wave-like dynamics in age-influential CpGs: an early-life module, a midlife transition, and late-life remodeling, with distinct timings by sex. These epigenetic waves cohere with non-linear, multi-omic “aging waves” reported in proteomics and longitudinal omics. SHAP further enabled interpretable CpG attribution, revealing structured, sex-specific aging phases: early-life male clocks involved developmental pathways, while female clocks emphasized cytoskeletal regulation; late-life divergence included immune activation in males and transcriptional remodeling in females. Our framework thus unites accuracy with mechanistic interpretability, revealing sex-specific windows when molecular aging reconfigures most rapidly.","manuscriptTitle":"Uncovering dynamics of age-related epigenetic changes with an interpretable deep-learning framework","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-18 02:47:10","doi":"10.21203/rs.3.rs-7614352/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-06T14:21:38+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-04T18:33:03+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-04T02:46:48+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-03T14:25:15+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-02T11:19:35+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-31T19:28:40+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"43180593982433451976353771357938486512","date":"2025-10-29T19:47:44+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"162125404958329782539047784380787827791","date":"2025-10-29T10:35:02+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-29T03:55:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"197579717571614564728112422480171394536","date":"2025-10-27T07:41:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"253362714892355748079613139243817007313","date":"2025-10-24T04:58:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"139842916590138215086403234740096824289","date":"2025-10-24T04:23:27+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"322428265460733070742925904536729083832","date":"2025-10-23T23:48:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"5843583690243598351014731759720771234","date":"2025-10-23T15:14:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"303595646795580818721945474839227320274","date":"2025-10-23T14:31:11+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-23T14:09:56+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-17T15:58:59+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-15T14:11:28+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Aging","date":"2025-09-14T18:55:19+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-aging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [npj Aging](https://www.nature.com/npjamd/)","snPcode":"41514","submissionUrl":"https://submission.springernature.com/new-submission/41514/3","title":"npj Aging","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"454d9b26-a11b-40a6-a8a0-b6e93ee7d4d3","owner":[],"postedDate":"September 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":54894548,"name":"Health sciences/Biomarkers"},{"id":54894549,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":54894550,"name":"Biological sciences/Genetics"}],"tags":[],"updatedAt":"2026-03-16T16:18:38+00:00","versionOfRecord":{"articleIdentity":"rs-7614352","link":"https://doi.org/10.1038/s41514-026-00358-w","journal":{"identity":"npj-aging","isVorOnly":false,"title":"npj Aging"},"publishedOn":"2026-03-13 15:59:38","publishedOnDateReadable":"March 13th, 2026"},"versionCreatedAt":"2025-09-18 02:47:10","video":"","vorDoi":"10.1038/s41514-026-00358-w","vorDoiUrl":"https://doi.org/10.1038/s41514-026-00358-w","workflowStages":[]},"version":"v1","identity":"rs-7614352","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7614352","identity":"rs-7614352","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}