DMCHMM

DMCHMM implements a Hidden Markov Model–based framework to identify differentially methylated CpG sites (DMCs) and regions (DMRs) from bisulfite sequencing data.

Key Features:

• Hidden Markov Model framework: Uses an HMM-based probabilistic model to infer methylation states and detect DMCs and DMRs.
• Per-sample profiling: Profiles methylation states independently for each sample to exploit intra-sample inter-CpG autocorrelation and allow region-specific hidden-state structures.
• Multiple hidden states: Supports modeling with multiple hidden states to capture nonstationary methylation patterns across genomic regions.
• Group comparisons and covariates: Accommodates comparisons across multiple groups and includes continuous covariates in the analysis.
• Missing-data handling: Handles settings with missing values within the methylation data.
• Robustness to sequencing challenges: Accounts for highly variable and low read depths, uneven CpG spacing, and strong local autocorrelation.
• Three-step analysis procedure: Implements model selection, state prediction, and statistical testing to produce DMC and DMR calls.

Scientific Applications:

• Differential methylation detection: Identification of DMCs and DMRs from bisulfite sequencing datasets.
• Low-coverage and uneven-spacing analyses: Analysis of bisulfite sequencing studies with low read coverage or heterogeneous CpG spacing.
• Complex study designs: Comparative analyses involving multiple groups and continuous covariates, including data with missing values.
• Cell-type–resolved methylation studies: Detection of biologically meaningful methylation differences in cell-type–resolved blood methylation data associated with exposures or phenotypic traits.
• Method benchmarking: Performance evaluation via simulation studies under scenarios with low coverage, heterogeneous CpG spacing, or complex covariate structures.

Methodology:

Applies a Hidden Markov Model fitted per sample and executes a three-step procedure of model selection, state prediction, and statistical testing while accounting for read-depth variability, uneven CpG spacing, nonstationary methylation, and local autocorrelation.