A hidden markov model for identifying differentially methylated sites in bisulfite sequencing data

Area of Science:

• Genomics and Bioinformatics
• Epigenetics and Gene Regulation
• Computational Biology and Statistical Genetics

Background:

• DNA methylation studies are crucial for understanding biological processes and diseases, but quantitative statistical analysis methods are limited.
• Existing methods for identifying differentially methylated CpG (DMC) sites or regions (DMR) struggle with bisulfite sequencing data challenges like variable read depth, autocorrelation, and uneven CpG distribution.
• Current tools often lack the flexibility to compare multiple groups, handle missing values, or incorporate continuous/multiple covariates, hindering comprehensive genomic association studies.

Purpose of the Study:

• To develop an efficient and flexible statistical method for identifying differentially methylated CpG (DMC) sites and regions (DMR) in DNA methylation studies.
• To overcome the limitations of existing methods in handling bisulfite sequencing data complexities and methodological constraints.
• To provide a tool that can effectively analyze multiple groups, manage missing data, and integrate continuous or multiple covariates for association studies.

Main Methods:

• Developed DMCHMM, a novel three-step (model selection, prediction, testing) method based on Hidden Markov Models (HMMs) for DMC identification.
• DMCHMM profiles methylation for each sample independently, leveraging inter-CpG autocorrelation within samples.
• The method incorporates flexibility through multiple hidden states, enhancing its applicability compared to previous HMM approaches.

Main Results:

• Simulation studies demonstrated that DMCHMM outperforms several competing methods in identifying differentially methylated sites.
• The method effectively addresses challenges such as variable read depths, regional methylation/autocorrelation changes, and uneven CpG distribution.
• DMCHMM successfully analyzed cell-separated blood methylation profiles, showcasing its practical utility.

Conclusions:

• DMCHMM offers an efficient and robust solution for identifying differentially methylated regions, overcoming key limitations of existing bisulfite sequencing data analysis tools.
• The method's flexibility in handling multiple groups, missing values, and covariates makes it a valuable asset for researchers investigating the relationship between methylation and various exposures or traits.
• DMCHMM represents a significant advancement in the statistical analysis of DNA methylation data, enhancing the understanding of epigenetic regulation in health and disease.