Detect differentially methylated regions using non-homogeneous hidden Markov model for bisulfite sequencing data

DNA methylation plays an important role in many biological processes and diseases. With the rise of the whole genome bisulfite sequencing technique, aberrant methylation patterns can now be detected by comparing paired normal and disease samples at the single nucleotide level. We develop a novel Bayesian method for detecting differentially methylated regions from paired bisulfite sequencing data, and implement it as a R package called BSDMR. Based on a non-homogeneous hidden Markov model, BSDMR provides a better modeling strategy for the spatial correlation between CpG sites and takes into consideration the relationship between methylation signals from normal and disease samples. Simulations show that BSDMR performs well even under low read depth and has a smaller false discovery rates than existing methods. We also apply BSDMR to the colon cancer data from Gene Expression Omnibus. The detected DMRs are well supported by existing biomedical literatures.

Automated summary

A novel Bayesian method called BSDMR was developed for detecting differentially methylated regions from paired bisulfite sequencing data, which provides a better modeling strategy for the spatial correlation between CpG sites and considers the relationship between methylation signals from normal and disease samples. Simulations showed that BSDMR performs well even under low read depth with smaller false discovery rates, and was validated in colon cancer data.