A robust computational framework for methylation age and disease-risk prediction based on pairwise learning.

Area of Science:

• Epigenetics
• Computational Biology
• Bioinformatics

Background:

• Conventional epigenetic clocks struggle with generalizability due to batch effects, limiting clinical use for aging assessment.
• Technical biases in DNA methylation data hinder accurate age and disease risk prediction.

Purpose of the Study:

• To introduce MAPLE, a robust computational framework for methylation age and disease-risk prediction.
• To address the limitations of existing epigenetic clocks in generalizability and clinical applicability.

Main Methods:

• MAPLE employs pairwise learning to analyze relative relationships between DNA methylation profiles.
• This approach identifies biological signals for aging and disease while mitigating technical data biases.

Main Results:

• MAPLE achieved a median absolute error of 1.6 years across 31 benchmark tests, outperforming five competing methods.
• It demonstrated strong performance in disease risk assessment, with AUCs of 0.97 for disease identification and 0.85 for pre-disease detection.

Conclusions:

• MAPLE offers a robust and generalizable approach to epigenetic age and aging-related disease risk prediction.
• The framework shows significant potential for clinical applications in aging assessment and disease management.