SOBDetector

SOBDetector detects strand orientation bias in variant calls by reanalyzing VCF records against original BAM files to distinguish likely artifacts from true variants for somatic and germline mutation filtration.

Key Features:

• VCF reanalysis: Reanalyzes mutations stored in Variant Call Format (VCF) files produced by variant callers.
• BAM-based read evaluation: Evaluates reads supporting alternate alleles using original binary alignment (BAM) files to assess strand orientation.
• Illumina-like paired-end support: Methodology is tailored for Illumina-like paired-end sequencing data.
• FFPE artifact assessment: Assesses the likelihood that observed variants are formalin-fixed paraffin-embedded (FFPE) induced artifacts.
• Artifact probability scoring: Estimates the probability that detected mutations are artifacts using a Bayesian logistic regression model and posterior distribution.
• Training data: Predictive model was trained on The Cancer Genome Atlas (TCGA) whole exomes.
• Compatibility: Designed to be compatible with common somatic single nucleotide variant (SNV)-calling pipelines.
• Implementation: Implemented in Java 1.8.

Scientific Applications:

• FFPE artifact filtration: Filtering formalin-induced artifacts from variant calls derived from FFPE samples.
• Somatic SNV filtration: Improving reliability of somatic single nucleotide variant calling in cancer genomics.
• Germline variant calling support: Assisting germline variant filtration by identifying strand-orientation–related artifacts.
• Variant-level quality control: Reassessing VCF-reported variants using BAM evidence to inform downstream analyses.
• Sequencing-study QC: Detecting strand-orientation bias in Illumina-like paired-end sequencing datasets.

Methodology:

Reanalyzes VCF entries by examining reads supporting alternate alleles in original BAM files to detect strand orientation bias; tailored for Illumina-like paired-end sequencing; applies a Bayesian logistic regression model using posterior distributions to score artifact probability; model trained on TCGA whole exomes.