ForestQC

ForestQC applies a random forest classifier together with conventional filtering to classify and filter genetic variants from next-generation sequencing (NGS) and whole-genome sequencing (WGS) data for quality control in variant analyses.

Key Features:

• Random forest classification: Uses a random forest model to predict the likelihood that a given variant is a false positive and to classify variants as high-quality or poor-quality.
• Sequencing quality metrics: Integrates sequencing depth, genotyping quality, and GC content as input features for variant quality assessment.
• Conventional filtering integration: Combines machine learning predictions with traditional filtering approaches to produce a comprehensive quality-control decision.
• Hardy–Weinberg Equilibrium handling: Computes HWE p-values when no HWE file is provided and calculates missing HWE p-values for sites absent from a supplied file.
• Benchmarking against VQSR: Evaluated using two distinct WGS datasets (related individuals from families and unrelated individuals) and reported superior performance compared to Variant Quality Score Recalibration (VQSR) in GATK.
• Scalability: Reported as efficient for application to large-scale sequencing datasets.

Scientific Applications:

• Variant quality control in NGS/WGS studies: Improves selection of reliable genetic variants for downstream analyses.
• Reduction of false positives: Identifies likely false-positive variant calls arising from NGS or variant-calling artifacts.
• Family-based and population-based studies: Applicable to both related-individual (family) WGS datasets and unrelated-individual WGS datasets.
• Alternative to VQSR: Serves as a comparative or replacement QC approach relative to GATK VQSR in variant filtering workflows.

Methodology:

Integrates sequencing depth, genotyping quality, and GC content as features; applies a random forest classifier to predict false-positive variants; combines machine learning predictions with conventional filtering approaches; computes HWE p-values when not provided or when missing for specific sites; evaluated on two WGS datasets (related and unrelated individuals) against GATK VQSR.