Jointly benchmarking small and structural variant calls with vcfdist.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Variant calling is crucial for genomic analysis, but current benchmarking tools often evaluate different variant types (SNPs, INDELs, SVs) separately.
• Existing methods face challenges in accurately assessing phased variant calls, particularly concerning phasing errors.

Purpose of the Study:

• To introduce vcfdist, a novel variant call benchmarking tool capable of jointly evaluating single-nucleotide polymorphisms (SNPs), insertions/deletions (INDELs), and structural variants (SVs) across the whole genome.
• To demonstrate the benefits of a joint evaluation approach for improving the accuracy and interpretability of variant call benchmarking.

Main Methods:

• Extension of the vcfdist tool to incorporate joint evaluation of SNPs, INDELs, and SVs.
• Application of the enhanced vcfdist tool to three independent datasets for comprehensive benchmarking.
• Analysis of phasing accuracy and error reduction through unified variant type evaluation.

Main Results:

• Joint evaluation uniformly reduced measured errors across all variant types: SNPs by 28.9%, INDELs by 19.3%, and SVs by 52.4%.
• vcfdist corrected a common flaw in phasing evaluations, reducing measured flip errors by over 50%.
• The tool demonstrated superior accuracy compared to previous methods and comparable performance to state-of-the-art approaches, with enhanced result interpretability.

Conclusions:

• Jointly benchmarking SNPs, INDELs, and SVs provides a more accurate and comprehensive assessment of variant call quality.
• vcfdist offers a significant advancement in variant call benchmarking, addressing limitations in previous tools and improving phasing evaluation.
• The enhanced interpretability of vcfdist results facilitates better understanding and utilization of genomic variant data.