Higher order methylation features for clustering and prediction in epigenomic studies

Area of Science:

• Epigenetics
• Computational Biology
• Genomics

Background:

• DNA methylation is a key epigenetic mark with an incompletely understood functional role.
• Quantitative associations between average DNA methylation and gene expression show poor correlations, especially outside CpG islands.

Purpose of the Study:

• To develop a machine learning approach to extract higher-order features from DNA methylation profiles.
• To investigate the relationship between these higher-order methylation features and gene expression.
• To identify distinct patterns of DNA methylation at promoter regions.

Main Methods:

• Utilized probabilistic machine learning to extract higher-order features from DNA methylation profiles.
• Developed a machine learning predictor of gene expression using these novel features.
• Employed clustering of promoter-proximal regions based on methylation patterns.

Main Results:

• Higher-order methylation features significantly improved gene expression prediction compared to average methylation levels.
• Identified five major patterns of DNA methylation across promoter regions in different cell lines.
• Provided evidence that DNA methylation beyond CpG islands may play a role in gene expression regulation.

Conclusions:

• Spatial correlations and the 'shape' of DNA methylation profiles are functionally relevant for gene expression.
• The developed methods allow for the quantification of these higher-order methylation features for downstream analysis.
• Methylation patterns beyond CpG islands offer new avenues for understanding epigenetic regulation.