biomvRCNS

biomvRCNS performs segmentation and annotation of continuous genomic signals from tiling arrays, microarrays, and next-generation sequencing (NGS) to detect transcripts, transcript variants, regions of deletion and amplification (copy number variations), chromatin-state changes, and DNA methylation ratios.

Key Features:

• Hidden Semi-Markov Model (HSMM): Implements an HSMM specifically tailored for diverse genomic profiles to segment continuous signal.
• Data type support: Processes tiling arrays, microarray, and next-generation sequencing (NGS) continuous signals and signal peaks.
• Flexible observation distributions: Supports various data/distribution models to serve as a general segmentation engine.
• Genomic-position-dependent sojourn: Incorporates genomic positions into the HSMM sojourn distribution to model biologically relevant segment lengths.
• Optional prior learning: Optionally utilizes prior learning from existing annotations or studies to inform segmentation.
• Feature detection: Detects transcripts, transcript variants, regulatory regions, copy number variations (regions of deletion and amplification), chromatin state changes, and DNA methylation ratios.
• Benchmarking: Evaluated by simulation benchmarking against state-of-the-art segmentation models with reported comparable or superior sensitivity and specificity.

Scientific Applications:

• Transcript and variant identification: Identification of transcripts and transcript variants from continuous signal data.
• Copy number variation detection: Detection of regions of deletion and amplification (copy number variations) from microarray or NGS data.
• Epigenetic state mapping: Mapping chromatin-state changes and quantifying DNA methylation ratios.
• Regulatory region annotation: Identification and annotation of regulatory regions for genome annotation from high-throughput data.
• Method benchmarking: Comparative assessment of segmentation algorithms via simulation benchmarking.

Methodology:

Uses a Hidden Semi-Markov Model (HSMM) with genomic-position-dependent sojourn distributions, supports multiple observation/data distributions, optionally incorporates priors learned from existing annotations or studies, and was assessed by simulation benchmarking against other segmentation models.