SaVor - A Reproducible Structural Variant Calling and Benchmarking Platform from Short-Read Data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Structural variations (SVs) are genomic differences >1 kilobase-pair (Kbp) arising from DNA repair, genome duplication, or transposable elements.
• Next-generation sequencing (NGS) has increased short-read genomic data, necessitating efficient SV detection methods.

Purpose of the Study:

• To introduce SaVor, a flexible and reproducible workflow for structural variation (SV) calling.
• To generate a consensus SV call-set from short-read sequencing data using user-defined merge parameters.

Main Methods:

• SaVor accepts single or multi-lane short-read paired-end Illumina data or BAM files.
• The workflow was tested on 1,165 *Arabidopsis thaliana* whole genome sequences.
• Performance was benchmarked against SVs identified using Lumpy.

Main Results:

• Intersection calls (SVs supported by 3 callers) achieved high precision (>0.91) but lower recall (<0.51).
• Union calls (SVs supported by at least 1 caller) showed high recall (>0.88) but lower precision (<0.57).
• Merge strategies impact the trade-off between precision and recall for SV call-sets.

Conclusions:

• SaVor provides a reproducible method for SV calling from short-read data.
• Downstream analysis requires careful consideration of precision-recall trade-offs based on the chosen merge strategy.
• SaVor is an open-source Snakemake pipeline available on GitHub.