A novel strategy for forensic age prediction by DNA methylation and support vector regression model

High deviations resulting from prediction model, gender and population difference have limited age estimation application of DNA methylation markers. Here we identified 2,957 novel age-associated DNA methylation sites (P < 0.01 and R(2) > 0.5) in blood of eight pairs of Chinese Han female monozygotic twins. Among them, nine novel sites (false discovery rate < 0.01), along with three other reported sites, were further validated in 49 unrelated female volunteers with ages of 20-80 years by Sequenom Massarray. A total of 95 CpGs were covered in the PCR products and 11 of them were built the age prediction models. After comparing four different models including, multivariate linear regression, multivariate nonlinear regression, back propagation neural network and support vector regression, SVR was identified as the most robust model with the least mean absolute deviation from real chronological age (2.8 years) and an average accuracy of 4.7 years predicted by only six loci from the 11 loci, as well as an less cross-validated error compared with linear regression model. Our novel strategy provides an accurate measurement that is highly useful in estimating the individual age in forensic practice as well as in tracking the aging process in other related applications.