ERDS

ERDS detects copy number variants (CNVs) in human genomes from next-generation sequencing (NGS) short-read data by integrating read depth and single-nucleotide variant information using paired Hidden Markov Models (PHMMs).

Key Features:

• Paired Hidden Markov Models (PHMMs): Models expected distribution of read depth from short reads and the presence of heterozygous sites to infer copy number states.
• Integration of read depth and SNV data: Combines read depth signals with single-nucleotide variant (heterozygous site) information to improve CNV calls.
• High sensitivity and specificity: Demonstrates high sensitivity and specificity for CNV detection in high-coverage genomes.
• Performance vs Genome STRiP: A comparative study (PMID: 22939633) reports comparable or better performance than Genome STRiP for common CNVs in high-coverage genomes.
• Rare CNV detection: Outperforms other methods in detecting rare CNVs under similar conditions.
• Handling amplified regions: Handles both unique and highly amplified genomic regions without requiring separate alignments for CNV and other variant calling.
• Computational efficiency: Optimized for computational efficiency on high-coverage whole-genome sequencing datasets.
• Exome limitation: Not suitable for whole exome sequencing data.

Scientific Applications:

• CNV detection in high-coverage WGS: Identification of copy number variants in human whole-genome sequencing (short-read NGS) datasets.
• Rare variant discovery: Detection and characterization of rare copy number variants in population or clinical sequencing projects.
• Analysis of amplified regions: Calling CNVs in unique and highly amplified genomic regions without separate alignments.
• Method benchmarking: Comparative evaluation of CNV callers, including benchmarking against Genome STRiP (PMID: 22939633).
• Not for exome studies: Explicitly indicates it should not be applied to whole-exome sequencing analyses.

Methodology:

Uses paired Hidden Markov Models (PHMMs) that model expected read-depth distributions from short reads and leverage heterozygous site information, integrating read depth analysis with single-nucleotide variant data and operating without separate alignments for CNV versus other variant calls.