CNV-JACG

CNV-JACG assesses and genotypes copy number variations (CNVs) from paired-end whole genome sequencing (WGS) data to evaluate CNV detection accuracy and genotypic attributes.

Key Features:

• Random forest model: Uses an ensemble learning random forest classifier for CNV assessment and genotyping.
• 21 distinctive features: Trains on 21 features that characterize CNV regions and their breakpoints.
• Breakpoint characterization: Incorporates features specific to CNV breakpoints as well as CNV region properties.
• Paired-end WGS input: Operates on paired-end whole genome sequencing data.
• Training datasets: Leverages data from the 1000 Genomes Project, Genome in a Bottle Consortium, Human Genome Structural Variation Consortium, and in-house technical replicates for model training and validation.
• Small CNV sensitivity: Demonstrates superior sensitivity for small CNVs (≤1 kb) compared to alternative genotyping methods.
• Mendelian inconsistency reduction: Reduces Mendelian inconsistencies within trios in comparative analyses.
• Technical-replicate concordance: Improves concordance between technical replicates.
• Comparative benchmarking: Evaluated against the genotyping method SV² for performance comparison.

Scientific Applications:

• CNV genotyping: Assigns genotypes to detected CNVs from WGS data.
• Accuracy assessment: Evaluates the accuracy of CNV calls and reduces false positives/negatives.
• Small-variant discovery: Enhances detection and validation of small CNVs (≤1 kb).
• Trio-based analyses: Identifies and reduces Mendelian inconsistencies in family-based studies.
• Technical reproducibility: Assesses concordance across technical replicates for sequencing experiments.
• Disease-related CNV studies: Supports identification of CNV contributors to human disease and investigations of CNV-linked heritability.

Methodology:

Train a random forest on 21 features describing CNV regions and breakpoints using labeled data from the 1000 Genomes Project, Genome in a Bottle Consortium, Human Genome Structural Variation Consortium, and in-house technical replicates; validate on paired-end WGS by comparing genotyping performance and metrics (Mendelian inconsistencies, technical-replicate concordance) against SV².