hiHMM

hiHMM infers chromatin state maps across multiple genomes using a hierarchically linked infinite Hidden Markov Model and genome-wide histone modification data to enable comparative epigenomic analysis.

Key Features:

• Bayesian Non-Parametric Framework: Employs a Bayesian non-parametric approach (hierarchically linked infinite HMM) that models chromatin states without predefining the number of states.
• Hierarchical Structure: Integrates multiple levels of biological information to produce consistent chromatin state definitions across different genomes.
• Joint Inference Across Multiple Genomes: Simultaneously analyzes and infers chromatin states in multiple genomes, supporting comparisons across species, cell types, and developmental stages.
• Adjustment for Species-Specific Biases: Accounts for variations in genome size, signal enrichment strength, and histone modification co-occurrence patterns to improve cross-genome comparisons.
• Input Data: Leverages genome-wide histone modification data as the primary signal for chromatin state inference.
• Validation: Method validation reported on synthetic datasets and real-world datasets from the modENCODE project.

Scientific Applications:

• Regulatory Element Definition: Maps chromatin states genome-wide to identify regulatory elements and infer their regulatory activities.
• Comparative Epigenomics: Enables cross-species and cross-condition analyses to investigate conserved epigenetic features and evolutionary differences.
• Developmental Biology: Supports analysis of chromatin dynamics across developmental stages to study changes in gene regulation over time.

Methodology:

Uses a hierarchical Bayesian non-parametric model extending hidden Markov models (the hierarchically linked infinite HMM), integrates multiple datasets for joint inference of chromatin states across genomes while accounting for shared and unique features, and has been validated on synthetic and modENCODE datasets.