DNA methylation-based forensic age prediction using artificial neural networks and next generation sequencing

Area of Science:

• Forensic Science
• Epigenetics
• Genomics

Background:

• Estimating donor age from biological material is crucial in forensic investigations.
• Aging involves molecular modifications, including epigenetic patterns like DNA methylation.
• Accurate age prediction models are needed for forensic applications.

Purpose of the Study:

• To develop an accurate model for predicting chronological age using age-specific DNA methylation patterns from whole blood.
• To evaluate the performance of machine learning models for age prediction.
• To assess the applicability of the developed model in non-blood samples and explore a next-generation sequencing (NGS)-based method.

Main Methods:

• Selected 45 age-associated CpG sites based on previous studies.
• Analyzed publicly available methylation data from 1156 whole blood samples (Illumina platforms).
• Applied stepwise regression and a generalized regression neural network (GRNN) model for age prediction. Developed an NGS-based method for methylation status quantification.

Main Results:

• A GRNN model using 16 CpG sites achieved high accuracy (R²=0.96, MAE=4.4 years) on a blind test set.
• The model demonstrated similar accuracy in independent cohorts (twins, disease states) and saliva samples (R²=0.96, MAE=4.0 years).
• An NGS-based method was developed, showing potential but requiring further optimization for accuracy (MAE=7.5 years).

Conclusions:

• Machine learning models based on DNA methylation offer a powerful tool for accurate chronological age prediction.
• The developed model is applicable to both blood and saliva samples, enhancing its forensic utility.
• Further optimization of NGS-based methods is needed to improve accuracy and reproducibility for forensic age estimation.