Abstract:DNA methylation is a dynamic process that adds methyl groups to DNA molecules, which plays an important role in gene expression regulation, genome stability, and aging. Epigenetic clocks developed over the past decade have used DNA methylation values at specific CpG sites combined with factors such as environmental influences and lifestyle factors to predict biological age with remarkable accuracy. However, these epigenetic clocks, built using conventional penalized regression models, have several limitations: (1) they assume consistent linear changes in methylation with age, even though methylation patterns may vary at different ages; (2) the selected CpGs in these clocks often lack specific biological significance; and (3) there is minimal overlap among the selected CpGs across different clocks, suggesting that only a small subset of a larger group of age-correlated CpGs is used to build an effective clock. Using data from 4,899 samples across 22 GEO datasets, we first analyzed the methylation patterns of 1,868 CpGs from 9 widely used epigenetic clocks and observed the expected consistently linear patterns in a subcluster. Next, we applied our own CpG selection method, which does not assume a lifelong linear correlation between DNA methylation and age. we divided the lifespan into overlapping age windows - [0, 20], [5, 25], [10, 30] ..., [55, 75], [60, 80] - and identified CpGs that showed strong correlations with age within each specific window. Most CpGs were selected in young and old age windows, while few were selected in the middle-aged windows, indicating fast DNA methylation changes during child development and aging. Among the 12,903 unique CpGs selected within any of the windows, we performed clustering and uncovered from main patterns: (1) increase before 20 years old, (2) decrease before 20 years old, (3) increase before 20 and then decrease after 60, and (4) decrease before 20 and then increase after 60. Comparing this to the clock CpGs, both sets contained CpGs that decreased during adolescence. When examining gender differences, we noted that female samples had notably slowed methylation changes in old age windows compared to male samples, possibly due to hormonal changes during menopause. These findings suggest that age-related methylation changes are more complex than previously thought, with implications for refining epigenetic clocks and redefining the way researchers study aging. Future research should explore the biological mechanisms behind these non-linear and sex-specific patterns.

Age Prediction Using DNA Methylation Heterogeneity Metrics

A Comprehensive Assessment of Methylation-Based Age Prediction Methods

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

A Path to Age Estimation through DNA Methylation

Epigenetic‐based age prediction in blood samples: Model development

A High‐throughput Targeted Bisulfite Sequencing‐based Analysis for Epigenetic Age Quantification and Monitoring

Assessment of DNA methylation from a single genomic region of ELOV2 is sufficient to predict chronological age

Epigenetic Predictor of Age

DNA methylation age of human tissues and cell types

DNA methylation markers to assess biological age

Age Prediction based on Blood DNA Methylation

Many chronological aging clocks can be found throughout the epigenome: Implications for quantifying biological aging

Epigenetic age prediction drifts resulting from next-generation methylation arrays

Age-Related Patterns of DNA Methylation Changes

Time is encoded by methylation changes at clustered CpG sites

DNA Methylation Clocks and Their Predictive Capacity for Aging Phenotypes and Healthspan

DNA methylation aging clocks: challenges and recommendations

Universal DNA methylation age across mammalian tissues

R Methylcipher: A Methylation Clock Investigational Package for Hypothesis-Driven Evaluation & Research

DeepMAge: A Methylation Aging Clock Developed with Deep Learning

Studies of hTERT DNA methylation assays on the human age prediction