BayesAge: A maximum likelihood algorithm to predict epigenetic age.

Area of Science:

• Epigenetics
• Bioinformatics
• Computational Biology

Background:

• DNA methylation patterns change predictably with age, serving as a biomarker for aging.
• Existing epigenetic clocks face challenges with missing data, count-based bisulfite sequencing, and interpretation.
• Penalized regression models are common but have limitations in handling complex methylation data.

Purpose of the Study:

• To introduce BayesAge, an advanced method for epigenetic age prediction using bulk bisulfite sequencing data.
• To address limitations of current epigenetic clock methodologies, including data handling and interpretability.
• To improve the accuracy and reliability of age estimation from DNA methylation patterns.

Main Methods:

• BayesAge extends the scAge methodology using maximum likelihood estimation (MLE) for age inference.
• It models count data with binomial distributions and uses LOWESS smoothing for non-linear methylation-age dynamics.
• The method is specifically designed for bulk bisulfite sequencing datasets.

Main Results:

• BayesAge shows superior performance over scAge and penalized regression methods.
• Age residuals from BayesAge exhibit no age association, indicating less bias.
• The method provides error bounds for age inference and achieves a higher coefficient of determination on down-sampled data.

Conclusions:

• BayesAge offers a significant advancement in epigenetic age prediction accuracy and interpretability.
• It effectively handles count-based data and non-linear age dynamics in bulk bisulfite sequencing.
• The ability to estimate error bounds enhances the reliability of epigenetic age estimation in aging studies.