A common epigenetic clock from childhood to old age

Area of Science:

• Forensic Science
• Genetics
• Biotechnology

Background:

• Forensic age estimation is a crucial DNA intelligence tool for criminal investigations, mass disaster identification, and legal disputes.
• Existing DNA methylation-based age prediction models are often limited to adult age ranges and primarily use blood samples.

Purpose of the Study:

• To develop and validate an extended age prediction model using DNA methylation data from a wide age range of Spanish individuals.
• To compare the accuracy of different statistical methods, including quantile regression (QR), quantile regression neural network (QRNN), and quantile regression support vector machine (QRSVM), for age prediction.

Main Methods:

• DNA methylation levels were analyzed using Agena Bioscience EpiTYPER® technology on 895 Spanish blood samples (ages 2-104).
• Seven CpG sites across seven genomic regions (ELOVL2, ASPA, PDE4C, FHL2, CCDC102B, MIR29B2CHG, chr16:85395429) were targeted.
• Three statistical methods (QR, QRNN, QRSVM) were employed and validated on an independent set of 152 samples.

Main Results:

• The QRNN and QRSVM methods demonstrated the highest accuracy, with a mean absolute prediction error (MAE) of ±3.36 and ±3.41, respectively.
• Validation on an independent dataset yielded similar accuracies (MAE: ±3.32 for QRNN, ±3.45 for QRSVM).
• Quantile regression methods provide age-dependent prediction intervals, adapting error estimates to the predicted age.

Conclusions:

• The developed extended age prediction model, utilizing DNA methylation in blood, is accurate across a broad age spectrum (2-104 years).
• QRNN and QRSVM are highly effective statistical approaches for forensic age estimation.
• A minimum training sample size of six samples per year-of-age is recommended for robust model development.