Please wait a minute...
 Home  About the Journal Editorial Board Aims & Scope Peer Review Policy Subscription Contact us
 
Early Edition  //  Current Issue  //  Open Special Issues  //  Archives  //  Most Read  //  Most Downloaded  //  Most Cited
Aging and disease
Orginal Article |
DeepMAge: A Methylation Aging Clock Developed with Deep Learning
Fedor Galkin1,2, Polina Mamoshina1, Kirill Kochetov1, Denis Sidorenko3, Alex Zhavoronkov1,3,4,*
1Deep Longevity Limited, Hong Kong.
2Integrative Genomics of Ageing Group, Institute of Ageing and Chronic Disease, University of Liverpool, Liverpool, UK.
3Insilico Medicine Hong Kong Limited, Hong Kong Science and Technology Park, Hong Kong.
4Buck Institute for Research on Aging, Novato, CA, USA.
Download: PDF(745 KB)   HTML
Export: BibTeX | EndNote | Reference Manager | ProCite | RefWorks    
Abstract  

DNA methylation aging clocks have become an invaluable tool in biogerontology research since their inception in 2013. Today, a variety of machine learning approaches have been tested for the purpose of predicting human age based on molecular-level features. Among these, deep learning, or neural networks, is an especially promising approach that has been used to construct accurate clocks using blood biochemistry, transcriptomics, and microbiomics data—feats unachieved by other algorithms. In this article, we explore how deep learning performs in a DNA methylation setting and compare it to the current industry standard—the 353 CpG clock published in 2013. The aging clock we are presenting (DeepMAge) is a neural network regressor trained on 4,930 blood DNA methylation profiles from 17 studies. Its absolute median error was 2.77 years in an independent verification set of 1,293 samples from 15 studies. DeepMAge shows biological relevance by assigning a higher predicted age to people with various health-related conditions, such as ovarian cancer, irritable bowel diseases, and multiple sclerosis.

Keywords aging      DNA methylation      epigenetics      artificial intelligence     
Corresponding Authors: Alex Zhavoronkov   
About author:

These authors contributed equally to this work.

Just Accepted Date: 04 December 2020  
Service
E-mail this article
E-mail Alert
RSS
Articles by authors
Fedor Galkin
Polina Mamoshina
Kirill Kochetov
Denis Sidorenko
Alex Zhavoronkov
Cite this article:   
Fedor Galkin,Polina Mamoshina,Kirill Kochetov, et al. DeepMAge: A Methylation Aging Clock Developed with Deep Learning[J]. Aging and disease, 10.14336/AD.2020.1202
URL:  
http://www.aginganddisease.org/EN/10.14336/AD.2020.1202     OR
CVHealthy verificationCase trainingCase verification
MedAE, years2.242.773.294.18
MAE, years3.213.804.745.08
R20.960.930.880.82
Pearson’s r0.980.970.940.94
RMSE, years4.555.447.516.24
N4,9301,2931,093439
Table 1  Accuracy metrics for DeepMAge. The accuracy achieved in cross-validation (CV column, MedAE = 2.24 years) was only slightly reduced during verification (healthy verification column, MedAE = 2.77 years). Accuracy declined in the samples with various health-related conditions (case verification column, MedAE = 4.35 years).
Figure 1.  Scatter plot of DeepMAge predictions in 4 data cohorts”. DeepMAge accurately predicted the chronological age of healthy people from the training set (A), healthy people from the verification set (B), and remained accurate in the aggregations of case cohorts from the studies included in the training set (C) and the verification set (D). Scatter plot in panel A shows the per-fold predictions obtained during CV, and the other panels show the predictions by the final model. MedAE = Median absolute error measured in years, N = Number of samples in a corresponding cohort (see Supplementary Figures 1-3 for a more detailed visualization).
SetError, yearsAbsolute Error, yearsN
Years(20-45)(45-55)(55-65)(65-75)(20-75)(20-45)(45-55)(55-65)(65-75)(20-75)
VerificationMale+0.48-2.50-1.46*-4.76*-0.87*+2.97+4.04+3.98+6.04*+3.68574
Female+0.23-3.58-0.06*-1.78*-0.12*+3.24+4.48+3.50+4.13*+3.40494
N707621631361068707621631361068
CVMale+0.62+2.14*+0.62*+0.81*0.97*+2.84+3.80+4.00+4.89+3.531452
Female+0.65+0.41*-0.54 *-2.17*-0.34*+2.76+3.59+3.77+4.58+3.592058
N1323670897620351013236708976203510
Table 2  DeepMAge prediction errors are not significantly different for younger males and females. Sex-related differences in age prediction for older adults are inconsistent between the CV and the verification sets.
Figure 2.  The DeepMAge prediction age distribution in the verification set closely resembled the real age distribution. Distributions were obtained using Gaussian kernel with 0.3σ bandwidth, where σ is the standard deviation of the age values.
GEO IDMean error in controlMean error in casesp-value (MW)p-value (random MW)N controlN caseN totalDeepMAge sampleCase cohort description
GSE53740*-0.37+0.632.70E-21.50E-1197186383TrainingNeurodegenerative tauopathy
GSE19711*-2.97-1.279.84E-64.39E-1272264536TrainingOvarian cancer
GSE77696+4.43+3.961.31E-15.29E-2117261378TrainingHIV
GSE106648*-1.84+0.262.17E-82.52E-1139140279TrainingMultiple sclerosis
GSE67530-2.66-1.631.12E-11.01E-110539144TrainingAcute respiratory distress syndrome
GSE525880.671.191.71E-14.84E-1582987TrainingDown syndrome
GSE97362*1.24-4.042.05E-39.30E-283150233TrainingCHARGE / Kabuki syndrome
GSE846240.540.734.39E-19.87E-2242448TrainingKawasaki disease
GSE1126964.244.563.44E-11.89E-16612VerificationMyasthenia gravis
GSE1021771.991.914.94E-12.38E-1181836VerificationMaternal gestational diabetes
GSE87582-9.59-3.791.08E-12.82E-112021VerificationHIV
GSE107737*-1.983.663.63E-31.56E-1121224VerificationCongenital
hypopituitarism
GSE87640*-0.201.031.24E-33.57E-184156240VerificationInflammatory bowel diseases
GSE99624-1.58-3.996.43E-23.76E-1163248VerificationOstheoporosis
Table 3  Five diseases (including ovarian cancer and multiple sclerosis) were associated with significantly higher age predictions (p-value (MW) < 0.05).
Figure 3.  DeepMAge, but not the 353 CpG clock, predicted donors with IBD (GEO study accession GSE87640) to be on average 1.23 years older than the healthy donors from the same study (p-value = 1.24E­3). Outliers outside the (-20; +20) prediction error window were removed from the image; The box is formed by the interquartile range with the median marked inside it. Whiskers protrude no farther than 1.5 times the interquartile range. GEO = Gene Expression Omnibus; IBD = Inflammatory bowel disease; N= Number of samples in a corresponding cohort.
Figure 4.  The DeepMAge clock shares 122 CpGs with the 353 CpG clock and seven CpGs with the 71 CpG clock. The latter two were published in 2013.
GEO IDMedAE, yearsPearson’s rNAge range, yearsMale ratio, %
DeepMAge353 CpGDeepMAge353 CpG
GSE107459 **1.633.430.790.6812718-350
GSE102177 *1.871.330.860.83184-1456
GSE34639 *1.920.220.890.88480-133
GSE105123 **2.062.870.470.3810719-2358
GSE61496 **2.143.420.970.9531030-7453
GSE87640 *2.523.020.860.878420-5862
GSE98876 **2.544.770.890.817126-69100
GSE79329 **2.633.740.920.893443-70100
GSE99624 **2.723.730.930.811649-8238
GSE107737 *3.033.620.340.461218-29100
GSE370083.742.260.810.819924-4537
GSE112696 *3.752.780.340.23622-2767
GSE59065 **4.355.010.950.9429522-8448
GSE103911 *6.966.140.850.766527-7771
GSE875829.596.41--160100
Average2.773.510.970.9312930-8452
Table 4  In seven out of 15 verification studies, DeepMAge performed better than the 353 CpG clock according to two quality metrics (MedAE and Pearson’s r).
[1] Schwartz PJ, O’Neill DV, Bentz ME, Brown A, Doyle BS, Liepa OC, et al. (2020). AI-enabled wargaming in the military decision making process. Proc SPIE, 11413:14130H.
[2] Rauschecker AM, Rudie JD, Xie L, Wang J, Duong MT, Botzolakis EJ, et al. (2020). Artificial intelligence system approaching neuroradiologist-level differential diagnosis accuracy at brain MRI. Radiology, 295:626-637.
[3] Knott KD, Seraphim A, Augusto JB, Xue H, Chacko L, Aung N, et al. (2020). The prognostic significance of quantitative myocardial perfusion. Circulation, 141:1282-1291.
[4] Petersen SE, Abdulkareem M, Leiner T (2019). Artificial intelligence will transform cardiac imaging—Opportunities and challenges. Front Cardiovasc Med, 6:133.
[5] Zhavoronkov A, Ivanenkov YA, Aliper A, Veselov MS, Aladinskiy VA, Aladinskaya AV, et al. (2019). Deep learning enables rapid identification of potent DDR1 kinase inhibitors. Nat Biotechnol, 37:1038-1040.
[6] Mamoshina P, Kochetov K, Putin E, Cortese F, Aliper A, Lee W-S, et al. (2018). Population specific biomarkers of human aging: A big data study using South Korean, Canadian, and Eastern European patient populations. J Gerontol A Biol Sci Med Sci, 73:1482-1490.
[7] Mamoshina P, Volosnikova M, Ozerov IV, Putin E, Skibina E, Cortese F, et al. (2018). Machine learning on human muscle transcriptomic data for biomarker discovery and tissue-specific drug target identification. Front Genet, 9:242.
[8] Galkin F, Mamoshina P, Aliper A, Putin E, Moskalev V, Gladyshev VN, et al. (2020). Human gut microbiome aging clock based on taxonomic profiling and deep learning. iScience, 23:101199.
[9] Galkin F, Mamoshina P, Aliper A, de Magalhães JP, Gladyshev VN, Zhavoronkov A (2020). Biohorology and biomarkers of aging: Current state-of-the-art, challenges and opportunities. Ageing Res Rev, 60:101050.
[10] Osborne B, Bakula D, Ben Ezra M, Dresen C, Hartmann E, Kristensen SM, et al. (2020). New methodologies in ageing research. Ageing Res Rev, 62:101094.
[11] Zhavoronkov A, Mamoshina P (2019). Deep aging clocks: The emergence of AI-based biomarkers of aging and longevity. Trends Pharmacol Sci, 40:546-549.
[12] Horvath S (2013). DNA methylation age of human tissues and cell types. Genome Biol, 14:R115.
[13] Hannum G, Guinney J, Zhao L, Zhang L, Hughes G, Sadda S, et al. (2013). Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol Cell, 49:359-367.
[14] Horvath S, Raj K (2018). DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nat Rev Genet, 19:371-384.
[15] Weidner C, Lin Q, Koch C, Eisele L, Beier F, Ziegler P, et al. (2014). Aging of blood can be tracked by DNA methylation changes at just three CpG sites. Genome Biol, 15:R24.
[16] Lin Q, Weidner CI, Costa IG, Marioni RE, Ferreira MRP, Deary IJ, et al. (2016). DNA methylation levels at individual age-associated CpG sites can be indicative for life expectancy. Aging, 8:394-401.
[17] Levine ME, Lu AT, Quach A, Chen BH, Assimes TL, Bandinelli S, et al. (2018). An epigenetic biomarker of aging for lifespan and healthspan. Aging, 10:573-591.
[18] Belsky DW, Moffitt TE, Cohen AA, Corcoran DL, Levine ME, Prinz JA, et al. (2017). Eleven telomere, epigenetic clock, and biomarker-composite quantifications of biological aging: Do they measure the same thing? Am J Epidemiol, 187:1220-1230.
[19] Zhavoronkov A, Mamoshina P, Vanhaelen Q, Scheibye-Knudsen M, Moskalev A, Aliper A (2019). Artificial intelligence for aging and longevity research: Recent advances and perspectives. Ageing Res Rev, 49:49-66.
[20] Du P, Kibbe WA, Lin SM (2008). lumi: A pipeline for processing Illumina microarray. Bioinformatics 24(13):1547-1548.
[21] Li Y, Chen C-Y, Wasserman WW (2016). Deep feature selection: Theory and application to identify enhancers and promoters. J Comput Biol, 23:322-336.
[22] Leray P, Gallinari P (1999). Feature selection with neural networks. Behaviormetrika, 26:145-166.
[23] Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salakhutdinov R (2014). Dropout: A simple way to prevent neural networks from overfitting. J Mach Learn Res, 15:1929-1958.
[24] Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, et al. (2000). Gene Ontology: Tool for the unification of biology. Nat Genet, 25:25-29.
[25] Williams GC (1957). Pleiotropy, natural selection, and the evolution of senescence. Evolution, 11:398-411.
[26] Zhang B, Gladyshev VN (2020). How can aging be reversed? Exploring rejuvenation from a damage-based perspective. Adv Genet, 1:e10025.
[27] Lu AT, Quach A, Wilson JG, Reiner AP, Aviv A, Raj K, et al. (2019). DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging, 11:303-327.
[1] Supplementary data Download
[1] Yan Zhao,Jun-Kun Zhan,Youshuo Liu. A Perspective on Roles Played by Immunosenescence in the Pathobiology of Alzheimer's Disease[J]. Aging and disease, 2020, 11(6): 1594-1607.
[2] Anwen Shao,Danfeng Lin,Lingling Wang,Sheng Tu,Cameron Lenahan,Jianmin Zhang. Oxidative Stress at the Crossroads of Aging, Stroke and Depression[J]. Aging and disease, 2020, 11(6): 1537-1566.
[3] Yan-hong Pan,Wei-peng Wu,Xing-dong Xiong. Circular RNAs: Promising Biomarkers for Age-related Diseases[J]. Aging and disease, 2020, 11(6): 1585-1593.
[4] Rohan Mishra,Bin Li. The Application of Artificial Intelligence in the Genetic Study of Alzheimer’s Disease[J]. Aging and disease, 2020, 11(6): 1567-1584.
[5] Preeti Gupta,Ryan Eyn Kidd Man,Eva K Fenwick,Amudha Aravindhan,Alfred TL Gan,Sahil Thakur,Bao Lin Pauline Soh,Joanne M Wood,Alex A Black,Angelique Chan,David Ng,Teoh Khim Hean,Edwin Goh,Chong Foong-Fong Mary,Jenny Loo,Ciaran Gerard Forde,Charumathi Sabanayagam,Ching-Yu Cheng,Tien Yin Wong,Ecosse L Lamoureux. Rationale and Methodology of The PopulatION HEalth and Eye Disease PRofile in Elderly Singaporeans Study [PIONEER][J]. Aging and disease, 2020, 11(6): 1444-1458.
[6] Rodrigo Urbina-Varela,María Inés Soto-Espinoza,Romina Vargas,Luis Quiñones,Andrea del Campo. Influence of BDNF Genetic Polymorphisms in the Pathophysiology of Aging-related Diseases[J]. Aging and disease, 2020, 11(6): 1513-1526.
[7] Alexey Moskalev,Ilia Stambler,Calogero Caruso. Innate and Adaptive Immunity in Aging and Longevity: The Foundation of Resilience[J]. Aging and disease, 2020, 11(6): 1363-1373.
[8] Zhang Xinmu,Li Lin. The Significance of 8-oxoGsn in Aging-Related Diseases[J]. Aging and disease, 2020, 11(5): 1329-1338.
[9] Li Xiaohua,He Shikun,Zhao Mingwei. An Updated Review of the Epigenetic Mechanism Underlying the Pathogenesis of Age-related Macular Degeneration[J]. Aging and disease, 2020, 11(5): 1219-1234.
[10] Li Tan,Thomas C Register,Raghunatha R Yammani. Age-Related Decline in Expression of Molecular Chaperones Induces Endoplasmic Reticulum Stress and Chondrocyte Apoptosis in Articular Cartilage[J]. Aging and disease, 2020, 11(5): 1091-1102.
[11] Lijun Zhao,Jianzhong Cao,Kexin Hu,Xiaodong He,Dou Yun,Tanjun Tong,Limin Han. Sirtuins and their Biological Relevance in Aging and Age-Related Diseases[J]. Aging and disease, 2020, 11(4): 927-945.
[12] Allen Caobi,Rajib Kumar Dutta,Luis D Garbinski,Maria Esteban-Lopez,Yasemin Ceyhan,Mickensone Andre,Marko Manevski,Chet Raj Ojha,Jessica Lapierre,Sneham Tiwari,Tiyash Parira,Nazira El-Hage. The Impact of CRISPR-Cas9 on Age-related Disorders: From Pathology to Therapy[J]. Aging and disease, 2020, 11(4): 895-915.
[13] Tian Li,Nan Mu,Yue Yin,Lu Yu,Heng Ma. Targeting AMP-Activated Protein Kinase in Aging-Related Cardiovascular Diseases[J]. Aging and disease, 2020, 11(4): 967-977.
[14] Asish K Ghosh. p300 in Cardiac Development and Accelerated Cardiac Aging[J]. Aging and disease, 2020, 11(4): 916-926.
[15] Alan R Hipkiss. COVID-19 and Senotherapeutics: Any Role for the Naturally-occurring Dipeptide Carnosine?[J]. Aging and disease, 2020, 11(4): 737-741.
Viewed
Full text


Abstract

Cited

  Shared   
Copyright © 2014 Aging and Disease, All Rights Reserved.
Address: Aging and Disease Editorial Office 3400 Camp Bowie Boulevard Fort Worth, TX76106 USA
Fax: (817) 735-0408 E-mail: editorial@aginganddisease.org
Powered by Beijing Magtech Co. Ltd