gVCF2CNV: a scalable pipeline for CNV detection from whole-genome sequencing data.

Conversion of gVCF files into Log R Ratio and B Allele Frequency signals compatible with established CNV callers

gVCF2CNV is a Nextflow pipeline that converts gVCF files, which are smaller than BAM/CRAM and routinely generated by variant-calling workflows, into Log R Ratio (LRR) and B Allele Frequency (BAF) signals compatible with established CNV callers. Applied to 12,509 individuals from the SPARK cohort, it generated signals at an average of 2.7 million SNV positions per individual and completed signal extraction in 4 hours using 192 CPUs. CNV calling with PennCNV and QuantiSNP identified candidate CNVs across a broad size range, with trio-based Mendelian precision of approximately 80% or higher for deletions of at least 30 kb and duplications of at least 5 kb. Application to 414,824 individuals from the All of Us cohort was completed in 96 hours, demonstrating feasibility at biobank scale; the code is available on GitHub.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Preprint, not peer reviewed. The tool primarily targets large research cohorts, where re-reading BAM files is prohibitive; trio-based Mendelian precision is an indirect metric, with no comparison to BAM-based CNV calling or to an orthogonal truth set. For diagnostics, precision is reported only above 30 kb for deletions, whereas small intragenic deletions are precisely what WGS is expected to deliver.

Analysis by Dr Thibaut Benquey