Segway

Segway annotates functional genomic regions by applying a dynamic Bayesian network to multiple tracks of chromatin data to produce genome-wide, single-base-pair resolution annotations.

Key Features:

• Dynamic Bayesian Network: Uses a dynamic Bayesian network (DBN) model to represent dependencies across genomic positions.
• Unsupervised Learning: Trains models in an unsupervised manner to discover chromatin-state patterns without labeled training data.
• Multi-track Integration: Simultaneously trains on multiple chromatin experiments including histone modifications, transcription-factor binding sites, and open chromatin data.
• Single-base-pair Resolution: Performs analysis at 1-bp resolution across the entire genome without downsampling.
• Missing-data Handling: Accommodates heterogeneous patterns of missing data across experiments during training and inference.
• ENCODE Integration: Converts extensive ENCODE chromatin datasets into discrete annotation maps summarizing regulatory regions genome-wide.
• Chromatin–Transcription Insights: Produces annotations that elucidate relationships between chromatin activity and RNA transcription and identify quiescent genome regions.
• Evolutionary Constraint Correlation: Generates non-coding regulatory annotations that correlate with mammalian evolutionary constraint.
• Disease-locus Reinterpretation: Reannotates disease-associated loci using chromatin landscape information to generate focused hypotheses.
• ChIP-seq Applicability: Supports interpretation of ChIP-seq experiments at high resolution.

Scientific Applications:

• Regulatory Element Annotation: Produces genome-wide maps of regulatory elements from ENCODE and other chromatin datasets.
• Chromatin Architecture Studies: Identifies chromatin states and their interplay with RNA transcription across cell types.
• Conservation Assessment: Provides annotations used to assess and correlate non-coding regulatory elements with mammalian evolutionary constraint.
• Disease-associated Locus Analysis: Reinterprets previously uncharacterized disease-associated loci in the context of chromatin landscape to generate testable hypotheses.
• ChIP-seq Data Interpretation: Enables high-resolution analysis and annotation of ChIP-seq-derived signals.

Methodology:

Segway applies a dynamic Bayesian network using unsupervised learning to simultaneously train on multiple chromatin tracks (histone modifications, transcription-factor binding sites, open chromatin), processes the genome at 1-bp resolution, accommodates heterogeneous missing data, and outputs discrete annotation maps.