Chronological Age Prediction: Developmental Evaluation of DNA Methylation-Based Machine Learning Models

Area of Science:

• Genetics and Genomics
• Biotechnology
• Computational Biology

Background:

• Epigenetic clocks, based on DNA methylation (DNAm), are crucial for estimating biological age and understanding aging mechanisms.
• Accurate age prediction has significant implications in forensics, justice, and social sciences.
• Developing population-specific epigenetic clocks is essential due to potential variations in DNAm patterns.

Purpose of the Study:

• To construct and validate a blood-based epigenetic clock for chronological age prediction in the Southern Han Chinese (CHS) population.
• To integrate machine learning algorithms for optimizing age prediction accuracy.
• To assess the performance of different machine learning models and identify the most robust predictor.

Main Methods:

• Correlation coefficient meta-analyses of 7,084 individuals identified five key age-associated genes (ELOVL2, C1orf132, TRIM59, FHL2, KLF14).
• DNA methylation profiles were generated using bisulfite targeted amplicon pyrosequencing (BTA-pseq) on 34 CpG sites from 240 CHS blood samples (ages 1-81).
• Four machine learning models (stepwise regression, SVR-eps, SVR-nu, RFR) were trained and evaluated using median absolute deviation (MAD).

Main Results:

• DNA methylation levels showed population specificity across different CpG sites.
• The random forest regression (RFR) model demonstrated the highest accuracy, with a MAD of 1.29 years in the CHS cohort.
• An optimized RFR model achieved a MAD of 1.15 years for ages 1-60, outperforming a meta-cohort range of 2.53-5.07 years.

Conclusions:

• A robust blood epigenetic clock was successfully developed for the Southern Han Chinese population using machine learning.
• The optimized RFR model provides a reliable tool for chronological age estimation in this demographic.
• This epigenetic clock has potential applications in age-related research and forensic contexts.