ChromHMM

ChromHMM applies a multivariate Hidden Markov Model to learn and annotate genome-wide chromatin states from combinatorial patterns of chromatin modifications, enabling segmentation of the genome into putative functional regulatory elements and interpretation of cell-type–specific regulatory landscapes and disease-associated loci.

Key Features:

• Multivariate Hidden Markov Model: Uses a multivariate HMM formulation to model combinatorial patterns of chromatin modifications across the genome.
• Emission and transition parameter learning: Learns probabilistic chromatin-state emission and transition parameters from aligned chromatin mark data.
• Input preprocessing: Provides automated conversion of raw chromatin mark data to presence/absence representations.
• Model inference: Performs model inference to derive chromatin-state parameters.
• Genome segmentation: Outputs genome-wide state assignments and segments the genome into chromatin states that represent putative regulatory elements.
• Visualization: Generates multi-scale visual summaries of chromatin states.
• Enrichment and correlation analysis: Correlates state annotations with large-scale external functional datasets and regulatory annotations to reveal characteristic enrichment patterns.
• Biological interpretation: Supports interpretation of cell-type–specific regulatory landscapes and disease-associated loci.

Scientific Applications:

• Regulatory element annotation: Segmenting the genome into putative promoters, enhancers, and other regulatory elements based on chromatin-state patterns.
• Cell-type–specific landscape analysis: Comparing chromatin-state architectures across diverse cell types to interpret cell-type–specific regulatory programs.
• Disease-locus interpretation: Annotating and prioritizing disease-associated loci by their chromatin-state context.
• Hypothesis generation via enrichment patterns: Revealing characteristic enrichments by correlating state annotations with external functional datasets and regulatory annotations.

Methodology:

Using a multivariate Hidden Markov Model, ChromHMM learns probabilistic emission and transition parameters from aligned chromatin mark data, converts raw chromatin mark data to presence/absence representations, performs model inference and state segmentation to produce genome-wide state assignments, and generates visual summaries; state annotations can be correlated with external functional datasets and regulatory annotations to reveal enrichment patterns.