Modeling Differentially Methylated Genomic Regions with Generalized Gaussian Process Regression

Cytonsine methylation is one of the major forms epigenetic modification and has been shown to play critical roles in various biological processes. The emergence of high throughput Bisulfite sequencing makes it possible to obtain a comprehensive view of methylation patterns over the entire genome with reasonable cost. Thus genome-wide comparisons of DNA methylation patterns between different genotypes or under different treatments have become increasingly common. Though a lot of information can be gained by studying differentially methylated individual cytocines, identifying genomic regions that are differentially methylated can potentially provide more information and be more reliable. However, compuational and statistical methods for locating differentially methylated regions are mostly ad hoc in nature, typically by clustering differentially methylated cytosines after position-wise tests. Here we propose a generalized Gaussian process regression based approach, which provides a formal framework for identifying differentially methylated regions. We illustrate the proposed method with a data set regarding Arabidopsis plants with different genotypes.