{"paper_id":"1f9feb9e-acc8-422f-a8b1-4416f2d044a3","body_text":"Age Prediction based on Blood DNA Methylation\nLIU Rui*, FAN Pu, ZHANG Jing, LIU Zhao, MI Hao-yuan\nSchool of Investigation, People's Public Security University of China, \nBeijing 100038, China\n* [Corresponding author] \nE-mail: liurui@ppsuc.edu.cn（LIU）\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nAbstract\n Objective In the judicial field, traditional DNA methylation age \nprediction models have low accuracy and poor stability. Additionally, the use of \nlinear regression models for detection is inefficient and costly. This study aims \nto utilize the prediction principles of the Support Vector Regression (SVR) \nmodel, based on preliminary laboratory data from blood DNA methylation \ndetection using the Illumina 850K chip. By selecting low-dimensional and \nhighly linear loci, we aim to establish a highly stable and accurate blood DNA \nmethylation age prediction model. Methods This research is based on Illumina \n850K chip technology. We conducted a literature review to select CpG sites and \nrelated primers, then employed SVR for model construction and age prediction. \nThe model was built on the Matlab2022a platform. Standard parameters were \nselected, and optimal values for C and g were determined using grid search and \ncross-validation methods. During data processing, numerical values were \nnormalized before calculation and de-normalized to obtain the predicted values. \nResults The constructed model achieved an R² of 0.91563 and a Mean Absolute \nError (MAE) of 2.77 years. This indicates that the prediction accuracy for blood \nsamples reached 91.56%, with an error of 2.77 years. Moreover, the accuracy of \nthe model's predictions decreases with increasing age.\nKeywords: forensic genetics; SVR; age estimation; DNA methylation\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n1.Introduction\nDNA methylation age estimation is based on the dynamic changes of \nDNA methylation patterns throughout the lifespan. By comparing an \nindividual's actual age with the level of DNA methylation in their \ngenome, predictive models can be established to accurately infer age. \nDNA methylation age estimation holds broad applications and scientific \nsignificance. It can assess physiological and health conditions, explore \nassociations between DNA methylation and aging, disease risks, and \nprovide references for personalized medicine and health management. \nFor instance, by comparing DNAm age with chronological age, potential \nhealth risks can be identified early, enabling appropriate preventive \nmeasures. Analyzing differences in DNAm age among different groups \ncan reveal the impact of environmental factors (such as lifestyle, diet, and \npollution) on biological age, thus informing health policies and \ninterventions.\nIn forensic practice, DNA methylation age estimation offers a novel \napproach for forensic medicine and individual identification. Analyzing \nDNA methylation levels to deduce biological age provides crucial \ninformation in legal contexts. Firstly, DNAm age can aid in criminal \ninvestigations by accurately determining the age of suspects using DNA \nmethylation analysis of extracted samples. Secondly, it facilitates age \ndetermination in cases involving unidentified individuals, such as \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nforensic examinations of bodies or child abduction cases. Thirdly, DNAm \nage can assist in defining criminal responsibility age, which varies \nbetween minors and adults in many legal systems.\nWhile DNA methylation holds significant potential in forensic \napplications, rigorous research and practical experience are essential to \nensure its scientific validity, stability, and accuracy in serving judicial \npractices objectively and fairly.\nWith the advancing field of epigenetics, the focus in human \ncharacterization studies is shifting towards single nucleotide \npolymorphisms (SNPs), alongside traditional markers like restriction \nfragment length polymorphisms (RFLP) and short tandem repeats (STR). \nDNA methylation, as a crucial epigenetic marker, complements these \nmarkers and plays a pivotal role in age estimation models based on its \ndynamic changes across the lifespan. Traditional methylation models \ntypically employ linear regression requiring multiple markers and \nsamples, thereby increasing complexity and costs. This study explores \nSVR modeling principles to predict age, utilizing fewer but stronger \nlinear markers, aiming to establish a highly stable and accurate blood \nDNA methylation age estimation model.\nBlood-based DNA methylation age estimation has been a prominent \nresearch area. Previous studies have utilized technologies like Illumina \nInfinium 450K chips and EpiTYPER systems to establish age prediction \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nmodels based on various CpG sites, demonstrating different levels of \nmean absolute deviation (MAD). The evolution of these models continues \nwith improvements in SVR models, aiming for robustness and \ncompatibility in age prediction from biological samples.\nIn this study, a comparative analysis between multiple linear \nregression (MLR) and SVR models using sample data suggests that SVR \nmodels provide higher repeatability, better fit, and greater compatibility \nfor developing accurate age prediction models based on blood DNA. \nThrough literature review and analysis, this research aims to select CpG \nsites common in saliva and blood DNA, analyze their methylation levels \nusing Illumina 850K chip technology, and establish a SVR regression \nmodel for blood-based DNA methylation age estimation.\n2. Methods\n2.1 Sample Collection\nThis study utilized Illumina 850K chip data from 80 unrelated \nindividual blood samples collected and tested by the Ministry of Public \nSecurity Appraisal Center. All samples were collected with informed \nconsent from the individuals and approved by the ethics committee of the \nMinistry of Public Security Appraisal Center. The age of blood samples \nranged from 20 to 59 years, with an average age of 39.08 years, as shown \nin Tab 1.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n2.2 Selection of CpG Sites\nBased on literature review and laboratory data from the Ministry of \nPublic Security Appraisal Center, 5 CpG sites potentially associated with \nage were initially selected as candidate sites for model training and \nresearch. These sites were numbered accordingly (as shown in Tab 2).\nTab.1 Blood sample distribution information\nAge groups  Sample size Average age\n20 ~ 29 20 23.55\n30 ~ 39 20 34.75\n40 ~ 49 20 44.45\n50 ~ 59 20 53.55\nTab.2 Initial screening site number information\nIdentification SITE\nCpG1 CG21572722\nCpG2 CG19283806\nCpG3 CG03607117\nCpG4 CG13552692\nCpG5 CG14692377\nCpG6 CG25410668\nCpG7 CG04875128\nCpG8 CG13552692\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n2.3 Pearson Correlation Analysis\nPearson correlation analysis is a statistical method to assess the \nlinear relationship and direction between two variables. In this \nexperiment, the \"cor()\" function in RStudio was used to analyze data \nfrom 80 samples across 5 candidate CpG sites detected by the 850K chip. \nThis analysis evaluated the correlation between methylation levels at each \ncandidate site and the corresponding age of the samples, yielding \ncorrelation coefficients (denoted as \"r\") and regression scatter plots.\n2.4 Selection and Optimization of Site Data\nTo meet the significant correlation requirement between variables \nusing Support Vector Machine (SVM), CpG sites with |r| > 0.5 from \nPearson correlation analysis were selected for subsequent SVR regression \nmodel training and construction. Outlier samples were identified and \nexcluded during data optimization to enhance prediction accuracy of the \nmodel. Normalization of selected sample data based on MATLAB \nR2022a's \"mapminmax('apply')\" function was performed to mitigate the \nimpact of differing magnitude levels on model performance and \ncomputation speed. The final prediction values were reverse-normalized \nusing \"mapminmax('reverse')\" to obtain age predictions in standard \nnumerical ranges.\n2.5 Construction of Age prediction Model\nSamples, after selection, were randomly shuffled with 80% used for \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\ntraining and 20% for testing the model. The SVR regression model was \nbuilt using MATLAB R2022a and the libsvm-3.34 toolkit, employing \ngrid search and cross-validation to optimize parameters C (regularization \nparameter) and γ (gamma). The \"svmtrain()\" function established the \nmodel for predicting age from methylation levels at CpG sites, while \n\"svmpredict()\" simulated age predictions based on methylation levels in \nthe test set. Visualization and mathematical analysis of model \nperformance were conducted through plotting and calculation of relevant \nerror metrics.\n2.6 Analysis of Model Accuracy\nThe accuracy of the age inference model was analyzed across four \nage groups (20-29, 30-39, 40-49, 50-59 years) using 76 selected blood \nsamples. Model performance indicators such as Mean Absolute Deviation \n(MAD) and accuracy were computed, with an error margin set at ±5 years \nconsidered acceptable.\nTab.3 R-value corresponding to candidate sites\nSite R\nCPG1 0.46\nCPG2 -0.54\nCPG3 0.61\nCPG4 -0.52\nCPG5 0.50\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n3. Results\n3.1 Pearson Correlation Analysis\nBased on Illumina 850K chip data from 80 blood samples, DNA \nmethylation levels at 8 CpG sites were correlated with sample age, as \ndepicted in scatter plots in Fig.1. The calculated Pearson correlation \ncoefficients are presented in Tab. 3. Among the 8 candidate sites, CpG2, \nCpG3, CpG4, CpG5, and CpG8 showed |r| values above 0.5, indicating a \nstrong positive correlation between methylation levels and sample age. \nCpG8, while more correlated than CpG5, exhibited more outliers, which \nwere unfavorable for SVR model construction. Therefore, CpG2, CpG3, \nCpG4, and CpG5 were chosen for the SVR model.\n3.2 Site Selection and Data Processing\nBased on Pearson correlation analysis, CpG2, CpG3, CpG4, CpG5, \nand CpG8 showed |r| values above 0.5, indicating a strong positive \ncorrelation between methylation levels and sample age. However, CpG8's \nnumerous outliers led to its exclusion from the study. Outlier samples \nwere identified and excluded from the training set, ensuring robust model \nperformance. Details are provided in the Tab.4.\nCPG6 0.37\nCPG7 0.47\nCPG8 -0.52\nTab.4 Abandoned blood sample information\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nSample number Age Age stage\nDZ-484 31 20~29\nJXH061 42 40~49\nJXH080 45 40~49\nJXH138 45 40~49\nJXS001 46 40~49\nL00341 50 40~49\nL00490 51 50~59\nL00398 57 50~59\nL00445\nL00445\n57\n57\n50~59\n50~59\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nFig.1 Regression scatter plot of DNA methylation levels at 5 CpG \nsites in blood samples and sample age\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n3.3 Construction of Blood DNA Methylation Age prediction \nModel\nUsing data from 70 selected samples, a model was built based on \nCpG2, CpG3, CpG4, and CpG5 sites. A mathematical formula for age \nprediction was derived using SVR regression, considering methylation \nlevels at these sites.\nF{x} = [x1, x2, x3，x4]，F{ω} = [ω1，ω2，ω3，ω4]\nyp  = F{ω} · F{x} + b\nYp  = yp × （N — M）+  M\n The model was trained and tested using MATLAB R2022a, \nyielding an MAE of 2.77 years and an accuracy of approximately 91.56% \nfor the test set. The model's performance was validated through \ncross-validation, ensuring its stability and accuracy. The selection of \ntraining model parameters and formula variable values is shown in Tab. \n5.\nYp = 40 × [ F{ω} · F{x} -0.40263 ] + 20\nTab.5 Training model parameters and formula variable values\nParameter (variable) Value Meaning\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nAccording to the calculation formula of relevant indicators, the \ntraining set R2 of the established SVR regression model is 0.90808, and \nthe testing set R2 is 0.88529. Therefore, the age inference model \nt 2 RBF kernel\nC 1\nRegularization parameter for \nregression models\ng 8\nParameters of the RBF kernel: \ngamma\ns 3\nTrain a model with epsilon-Support \nVector Regression \np 0.01 The epsilon value in the loss function\nω1 -0.9350 Estimate CpG2\nω2 0.8352 Estimate CpG3\nω3 -0.8754 Estimate CpG4\nω4 0.7378 Estimate CpG5\nb\n-0.4026\n3\nbias term in a regression model\nM 20\nsetting the left boundary of the \nsample interval\nN 60\nsetting the right boundary of the \nsample interva\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nconstructed based on the CpG2-CpG3-CpG4-CpG5 locus combination \ncan achieve an accuracy of about 89% for age prediction inference \nvalues, with a MAE value of 1.58 years for predicting age in the training \nset and actual age; The MAE value between the predicted age and the \nactual age in the test set is 2.57 years old. In the SVR regression model \nconstructed this time, the fitting line graph between the predicted age of \nthe training set and the true age is shown in Fig. 2, and the fitting line \ngraph between the predicted age of the testing set and the true age is \nshown in Fig. 3.\nFig.2 Fitting line graph between predicted age and actual age in the \ntraining set\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n图 3 测试集预测年龄与真实年龄的拟合折线图\nFig.3 Fitting line graph of predicted age and actual age in the test set\n3.4 Accuracy Analysis across Different Age Groups\nThe model's accuracy was assessed across four age groups using 60 \ntraining samples and 31 test samples. Performance metric ‘Accuracy’ are \npresented in Tab. 6 and Tab. 7 respectively. Despite decreasing Accuracy \nvalues with age group, indicating reduced predictive capability for older \nages, the model demonstrated robust accuracy within ±5 years.\nTab. 6: Accuracy metrics for age groups in the training set\nAge groups Accuracy Number of samples\n20-29 94.1% 17\n30-39 88.9% 18\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n4. Discussion\nRecent studies on DNA methylation age inference have increasingly \nfocused on using a small number of highly linear CpG sites for model \nconstruction. In this study, CpG2, CpG3, CpG4, and CpG5 showed \noptimal correlation (r-values) with sample age, facilitating SVR model \nconstruction. \nThrough iterative training and refinement, a robust SVR \nmathematical model was established, achieving a 89.28% accuracy rate \nwith a MAE of 2.77 years for age prediction. The model exhibited high \nstability and avoided overfitting, maintaining a self-check accuracy rate \nof 90%. However, as noted, MAD values increased with older age \ngroups, suggesting a decline in predictive accuracy for older individuals.\n40-49 84.6% 13\n50-59 83.3% 12\nTab. 7: Accuracy metrics for age groups in the test set\nAge groups Accuracy Number of samples\n20-29 100% 9\n30-39 100% 8\n40-49 75% 7\n50-59 57.1% 7\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nConflicts of interest\nThe authors declare that they have no known competing financial \ninterests or personal relationships that could have appeared to influence \nthe work reported in this paper.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\nReferences\n[1] Fraga MF, Ballestar E, Paz MF, Ropero S, Setien F, Ballestar ML, Heine-Suñer D, Cigudosa JC, \nUrioste M, Benitez J, Boix-Chornet M, Sanchez-Aguilera A, Ling C, Carlsson E, Poulsen P, Vaag \nA, Stephan Z, Spector TD, Wu YZ, Plass C, Esteller M. Epigenetic differences arise during the \nlifetime of monozygotic twins[J]. Proc Natl Acad Sci U S A, 2005, 102(30):10604-10609.\n[2] Christensen BC, Houseman EA, Marsit CJ, Zheng S, Wrensch MR, Wiemels JL, Nelson HH, \nKaragas MR, Padbury JF, Bueno R, Sugarbaker DJ, Yeh RF, Wiencke JK, Kelsey KT. Aging and \nenvironmental exposures alter tissue-specific DNA methylation dependent upon CpG island \ncontext[J]. PLoS Genet, 2009, 5(8):e1000602.\n[3]HORVATH S. Erratum to DNA methylation age of human tissues and cell types[J]. Genome Biol，\n2015，16：96.\n[4]Hannum G, Guinney J, Zhao L, Zhang L, Hughes G, Sadda S, Klotzle B, Bibikova M, Fan \nJB,Gao Y, Deconde R, Chen M, Rajapakse I, Friend S, Ideker T, Zhang K. Genome-wide \nmethylation profiles reveal quantitative views of human aging rates[J]. Mol Cell, \n2013,49(2):359-367. \n[5] Zbie ć-Piekarska R, Spólnicka M, Kupiec T, Parys-Proszek A, Makowska Ż, Pa łeczka A, \nKucharczyk K, Płoski R, Branicki W. Development of a forensically useful age prediction method \nbased on DNA methylation analysis[J]. Forensic Sci Int Genet,2015 ,17:173-179. \n[6] Horvath S. DNA methylation age of human tissues and cell types[J]. Genome Biology, 2013, \n14(10):R115.\n[7] Levine ME, Lu AT, Quach A, Chen BH, Assimes TL, Bandinelli S, Hou L, Baccarelli AA, \nStewart JD, Li Y, Whitsel EA, Wilson JG, Reiner AP, Aviv A, Lohman K, Liu Y, Ferrucci L, \nHorvath S. An epigenetic biomarker of aging for lifespan and healthspan[J]. Aging (Albany NY). \n2018,10(4):573-591.\n[8]LIU C，MARIONI R E，HEDMAN A K ，et al. A DNA methylation biomarker of alcohol \nconsumption[J]. Mol Psychiatry，2018，23（2）：422-433.\n[9]Hao T, Guo J, Liu J, Wang J, Liu Z, Cheng X, Li J, Ren J, Li Z, Yan J, Zhang G. Predicting \nhuman age by detecting DNA methylation status in hair[J]. Electrophoresis, 2021, 42(11): \n1255-1261.\n[10] Belsky DW, Moffitt TE, Cohen AA, Corcoran DL, Levine ME, Prinz JA, Schaefer J, Sugden K, \nWilliams B, Poulton R, Caspi A. Eleven telomere, epigenetic clock, and biomarker-composite \nquantifications of biological aging: do they measure the same thing?[J]. American Journal of \nEpidemiology, 2018, 187(6):1220-1230.\n[11]Zhang, Y., Wang, Y., & Shen, H. B. . DeepAge: a novel deep learning model for accurate age \nprediction based on DNA methylation. BMC genomics,2021, 22(1), 1-14.\n[12]Horvath, S., & Raj, K.DNA methylation-based biomarkers and the epigenetic clock theory of \nageing. Nature Reviews Genetics, 2021, 22(3), 173-185.\n[13] Smith BH, Campbell A, Linksted P, Fitzpatrick B, Jackson C, Kerr SM, Deary IJ, Macintyre \nDJ, Campbell H, McGilchrist M, Hocking LJ, Wisely L, Ford I, Lindsay RS, Morton R, Palmer CN. \nCohort profile: Generation Scotland: Scottish Family Health Study (GS: SFHS). The study, its \nparticipants and their potential for genetic research on health and illness[J]. International Journal of \nEpidemiology, 2013, 42(3):689-700.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n[14]Wang, L., Zhao, Y., Xu, Y., Xu, Z., Yao, Y., Huang, L., ... & Liu, Y. Effects of environmental \nexposures on DNA methylation-based age prediction: A study in a Chinese population. \nEnvironmental Pollution, 2023 292, 118419.\n[15]Li, Z., Zhang, W., Li, J., Zhou, L., & Wang, H.Cross-tissue DNA methylation age prediction \nbased on single-cell sequencing data. Bioinformatics, 2024,40(1), 122-130.\n[16]Zhu, Y., Huang, T., & Zhou, X. Identification of DNA methylation markers associated with \nlongevity using artificial intelligence. Aging Cell,2024,23(3), e13500.\n[17]Hannum G, Guinney J, Zhao L, Zhang L, Hughes G, Sadda S, Klotzle B, Bibikova M, Fan \nJB,Gao Y, Deconde R, Chen M, Rajapakse I, Friend S, Ideker T, Zhang K. Genome-wide \nmethylation profiles reveal quantitative views of human aging rates[J]. Mol Cell, \n2013,49(2):359-367. \n[18] Zbieć-Piekarska R, Spólnicka M, Kupiec T, Parys-Proszek A, Makowska Ż, Pałeczka A, \nKucharczyk K, Płoski R, Branicki W. Development of a forensically useful age prediction method based \non DNA methylation analysis[J]. Forensic Sci Int Genet,2015 ,17:173-179.\n[19]Xu C, Qu H, Wang G, Xie B, Shi Y, Yang Y, Zhao Z, Hu L, Fang X, Yan J, Feng L. A novel \nstrategy for forensic age prediction by DNA methylation and support vector regression model[J]. \nScientific reports, 2015, 5(1): 1-10.\n[20] XU C ， QU H ， WANG G ， et al. A novel strategy for forensic age prediction by DNA \nmethylation and support vector regression model[J]. Sci Rep，2015，5：17788.\n[21]Feng L, Peng F, Li S, Jiang L, Sun H, Ji A, Zeng C, Li C, Liu F. Systematic feature selection \nimproves accuracy of methylation-based forensic age estimation in Han Chinese males[J]. Forensic \nScience International: Genetics, 2018, 35: 38-45.\n[22] Pan C, Yi S, Xiao C, Huang Y, Chen X, Huang D. The evaluation of seven age-related CpGs \nfor forensic purpose in blood from Chinese Han population[J]. Forensic Sci Int \nGenet,2020 ,46:102251.\n[23]Smith, Z. D., & Meissner, A. DNA methylation: roles in mammalian development. Nature \nReviews Genetics,2013，14(3), 204–220.\n[24]Jones, P. A., & Takai, D.The r ole of DNA methylation in mammalian epigenetics. Science, \n2001,293(5532), 1068–1070.\n[25]Lu, A. T., Quach, A., Wilson, J. G., Reiner, A. P., Aviv, A., Raj, K., ... & Horvath, S. DNA \nmethylation GrimAge strongly predicts lifespan and healthspan. Aging,2020,12(10), 9711-9728.\n[26]Knight, A. K., Craig, J. M., Theda, C., Bækvad-Hansen, M., Bybjerg-Grauholm, J., Hansen, C. \nS., ... & Hollegaard, M. V. The epigenetic clock during human aging and the aging of stem cells and \ntheir niches. Nature Aging,2022, 2(4), 254-271.\n[27]Gao, X., Jia, M., Zhang, Y., Liu, H., Li, X., & Yang, X. Dynamic DNA methylation patterns in \nthe nervous system and their association with neurodegenerative diseases. Neuroscience & \nBiobehavioral Reviews,2024,128, 82-94.\n[28]Zhang, Z., Li, J., Yu, Y., & Li, D.. Application of DNA methylation in forensic science. \nForensic Sciences Research, 2020,5(4), 399-405.\n[29]Ballard, D., Giese, R. W., & Martello, S. DNA methylation age estimation in bloodstains: A \npowerful tool for forensic science. Forensic Science International: Genetics, 2020,48, 102340.\n[30]Freire-Aradas, A., Phillips, C., Lareu, M. V., & Carracedo, Á.Forensic individual age estimation \nwith DNA: From initial approaches to methylation tests. Forensic Science Reviews,2016, 28(2), \n111-126.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint \n\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted October 29, 2024. ; https://doi.org/10.1101/2024.10.25.620181doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}