A Genome-Wide Methylation Profiling of A Meta Cohort Dataset To Predict Biological Age Using Reproducing Kernel Hilbert Spaces and Bayesian Ridge Regression

Background and aim: the use of dna methylation to predict chronological age and aging rate is one potential application in many fields, including diseases prevention and treatment, forensics, and anti-aging medicine. although a large number of methylation markers have significant associations with age, most of the introduced age-prediction methods used a small number of the markers selected based on previously published studies or used datasets of methylation information from methylation array. methods: we implemented reproducing kernel hilbert spaces (rkhs) regression and ridge regression in a unified bayesian framework (brr) that make use of phenotypic and all methylation profiles simultaneously to predict chronological age. we used measurements at more than 450,000 cpg sites from the whole blood of a large cohort of 4,409 individuals with a wide age range of 10-101 years. evaluation of models fitted on all methylation profiles conducted using the adjusted and non-adjusted methylation measurements for cell heterogeneity. results: results showed that non-adjusted methylation presented significantly higher estimates of prediction accuracy compared to the adjusted methylation data, with a correlation between age and predicted age of 97.7% and a root-mean square error (rmse) of 3.54 years in non-adjusted versus correlation of 90.2% and rmse of 7.16 years in adjusted. reducing the number of predictors through ewas subset selection showed that the predictive power increased with a correlation of 98.3% and rmse 2.98 years in rkhs model. conducting epigenome wide association study using this huge data, we showed a significant distinct global methylation pattern over lifetime in cpg types with hypermethylation in the cpg islands and hypomethylation in the other cpg types including cpg shore, shelf and open sea (p < p < 5e-06). we also indicated that epigenetic drift is a widespread phenomenon where more than 97% of the age-associated methylation sites showed heteroscedasticity. conclusion: our results showed that the apparent methylomic aging rate (amar) had sex-specific pattern with increasing amar in women with age compared to men. according to these results, we concluded that brr and rkhs are powerful approaches to predict chronological age and dna methylation is more important to make differences in sex-specific amar.