Epigenetic Biomarkers for Age Estimation in Forensic Samples: A CpG-Site-Specific DNA Methylation Approach Using Machine Learning for Biological Age Prediction

Abstract

An important but difficult area of use in criminal investigations is forensic age estimation which is especially needed in cases where there is no documentation of the crime. The stabilized and heritable epigenetic alteration, DNA methylation, has become a prospective chronological age inferential biological clock. We applied bisulfite pyrosequencing to six CpG loci, namely, cg16867657 (ELOVL2), cg22736354 (SCGN), cg06493994 (FHL2), cg19283806 (C1orf132), cg17861230 (KLF14) and cg02228185 (TRIM59), on a cohort of 180 forens Mean absolute error (MAE) of 2.91 years, root mean squared error of 3.68 years, and an R2 coefficient of determination of 0.961 all indicated a better-training random forest regression model on these methylation beta-values, as compared to multiple linear regression, a support vector regression, and a gradient boosting equivalent. The model was tested cross-tissue-wise by validation between various forensic samples. The strongest predictors were found to be ELOVL2 and SCGN methylation with the analysis of feature importance. We have found a consistent, tissue-efficient system of prediction of age in a forensic context using a panel of DNA methylation, which has been checked in simulated case-work conditions.