A new robust AI/ML based model for accurate forensic age estimation using DNA methylation markers.

Abstract

CpG sites are regions of DNA where a cytosine nucleotide is followed by a guanine nucleotide in the 5' → 3' direction. Epigenetic markers based on methylation values at CpG sites are valuable for accurate age prediction and have become essential in forensic science, supporting criminal investigations and human identification. The present study identified 12 CpG sites from a collection of 476,366 CpG sites based on the following criteria: (a) CpG sites were retained if the Pearson correlation coefficient between the methylation values and the chronological age of the individual is greater than 0.85, and (b) if the mutual correlation coefficient between a pair of selected CpG sites is greater than 0.15, only one of them is retained. The identified CpG sites are associated with genes FHL2, ELOVL2, TRIM59, PCDHB1, KLF14, C1orf132, ACSS3, and CCDC102B. To ensure that the predictive accuracy is intrinsic to the selected CpG sites and not model dependent, the identified CpG sites were passed to three different Neural network models. All models achieved comparable accuracy across diverse populations, genders, and health conditions. The model's accuracy and reliability were validated through age predictions on independent datasets. By utilizing a minimal set of CpG sites, this approach offers a robust and efficient solution for forensic age estimation, significantly enhancing the precision and reliability of forensic investigations.