|
Abstract:
|
A research topic in DNA methylation is to detect differentially methylated regions (DMRs) between two groups of interest, such as normal versus tumor. Various methods have been developed for the detection of DMRs and/or differentially methylated cytosines (DMCs), which can be divided into two categories by the type of data: bisulfite-based sequencing (BS-seq) and microarray data. Besides disease status, covariates such as age and sex may also affect DNA methylation and thus need to be accounted for in differential methylation (DM) analysis. Another challenge lies in the fact that tumor tissues are a mixture of different cell types, which should be considered to avoid bias. In our work, a Bayesian curve credible bands approach was developed to detect DMRs for both BS-seq and microarray data under a unified framework, a feature that most of the other methods do not have. An R package, BCurve, was developed to implement the proposed model, which takes the spatial correlation of nearby CG sites, covariate effects, between sample variability and sample proportion into consideration. Our method was applied to simulated as well as real data, showing its superiority over existing methods.
|
