A blood DNA methylation age-prediction model based on support vector regression

Based on blood DNA methylation data obtained from the Illumina Methylation EPIC (850K) Bead Chip, this study aims to establish a highly stable and accurate age-prediction model has been developed using the principles of support vector regression (SVR) with a low-dimensional and high-correlation site selection approach. Four core CpG sites were selected through a systematic literature review and Pearson correlation analysis. The SVR model was developed in Matlab platform using the libsvm toolkit. To determine the optimal parameter combination, grid search combined with cross-validation was employed to optimize the penalty parameter C and the kernel parameter g. During data processing, DNA methylation β-values were normalized for model training and subsequently denormalized to obtain predicted age values. The results demonstrate that the constructed model achieved a coefficient of determination (R²) of 0.885 29 and a mean absolute deviation (MAD) of 2.57 years on the test set, indicating high overall predictive accuracy. Although the prediction error increased with age—leading to a slight decrease in accuracy for older age groups—the error remained within an acceptable range. This study demonstrates that the SVR model can achieve high-precision age estimation using only four CpG sites, demonstrating robust generalization capability and practical utility. This method provides effective technical support for the physiological characterization of criminal suspects in forensic practice.