A framework for genome-wide characterization of DNA methylation and variant effect prediction

DNA methylation plays a critical role in many essential biological processes, including transcriptional regulation, genomic imprinting, chromosomal maintenance, genomic stability and development, and has phenotypic and disease impacts. However, genome-wide characterization of methylation patterns across cell types and tissues remains a major challenge. Here, to address this, we present a framework that enables systematic genome-wide analysis of methylation and sequence-based prediction of context-specific methylation effects of variants. Our data-driven discovery of methylation classes reveals context-dependent roles for methylation across tissues, cell types and regions of the genome. The framework also includes Wreath, a deep learning-based sequence model specifically designed to accurately predict the context-dependent impact of any genetic variant on methylation. These developments leverage a compendium that harmonizes disparate methylation profiles from disparate studies including over 90 tissues and cell types. Using Wreath, we systematically characterized the impact of noncoding pathogenic variants on tissue-specific methylation, generating mechanistic hypotheses for individual regulatory variants. We make the full framework, including the compendium, the methylation classes and the model, available for use by the community. This study introduces a framework for integrating DNA methylation data across more than 90 tissues and cell types, characterizing genome-wide methylation patterns and predicting context-specific effects of genetic variants on DNA methylation and regulatory function using a deep learning model, Wreath.