NPSV: A simulation-driven approach to genotyping structural variants in whole-genome sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Structural variants (SVs) are crucial in disease etiology.
• Accurate genotyping of SVs from whole-genome sequencing data remains challenging.
• Existing SV genotypers struggle with variant and sample-specific biases.

Purpose of the Study:

• To introduce NPSV, a novel machine learning approach for genotyping known structural variants.
• To address biases in next-generation sequencing data for improved SV genotyping.
• To enhance the accuracy of structural variant detection in genomic analyses.

Main Methods:

• Developed NPSV, a machine learning-based genotyper.
• Utilized next-generation sequencing simulation to model genomic, sequencer, and alignment pipeline effects.
• Evaluated NPSV against existing genotypers on benchmark call sets.

Main Results:

• NPSV achieves state-of-the-art genotyping accuracy across diverse SV call sets, samples, and variant types.
• Demonstrated consistent or superior performance compared to existing methods.
• Successfully identified de novo SVs in trio contexts and showed robustness to breakpoint offsets.

Conclusions:

• Stand-alone SV genotyping is increasingly vital with growing SV databases and long-read sequencing data.
• NPSV offers a robust framework for accurate SV genotyping in various applications by simulating biases.
• The approach facilitates accurate genotyping of a wide range of SVs in targeted and genome-scale studies.