HaMMLET

HaMMLET performs Bayesian inference on Hidden Markov Models (HMMs) using a Fast Forward-Backward Gibbs sampler to detect and analyze copy number variants (CNVs) in array Comparative Genomic Hybridization (CGH) data.

Key Features:

• Bayesian HMM inference: Performs Bayesian inference on HMMs via a Fast Forward-Backward Gibbs sampler.
• Haar wavelet integration: Integrates the Haar wavelet transform with HMMs to enable data compression.
• Adaptive compression: Dynamically compresses data during each iteration based on current variance samples.
• Focused computation: Concentrates computational effort on chromosomal segments that are difficult to call.
• Block recomputation: Dynamically and adaptively recomputes consecutive blocks of observations likely to share a copy number.
• Automatic prior: Implements an effective automatic prior for the Bayesian model.
• Reduced runtime: Achieves significantly reduced computational time for Bayesian inference compared to standard Gibbs sampling approaches.

Scientific Applications:

• CNV detection in array CGH: Detection and analysis of genomic copy number variants (CNVs) in array Comparative Genomic Hybridization (CGH) experiments.
• Challenging region analysis: Enhanced calling accuracy for chromosomal segments that are challenging to analyze.
• Legacy data re-analysis and diagnostics: Re-analysis of legacy array CGH data collections and support for routine diagnostic CNV assessment.

Methodology:

Uses a Fast Forward-Backward Gibbs sampler for Bayesian inference on HMMs; applies the Haar wavelet transform to dynamically compress data based on current variance samples during each iteration; adaptively recomputes consecutive blocks of observations likely to share a copy number; and employs an automatic prior.