# ASVBM: Structural variant benchmarking with local joint analysis for multiple callsets

## TL;DR

ASVBM is a new benchmarking framework for structural variant detection that improves accuracy by jointly analyzing multiple variants.

## Contribution

ASVBM introduces latent positives and a joint analysis strategy to better match variants across callsets.

## Key findings

- Multiple small variants can be equivalent to a larger variant, improving benchmarking accuracy.
- ASVBM reduces false mismatches by jointly analyzing local variants in callsets.
- ASVBM was evaluated using real WGS datasets and six state-of-the-art pipelines.

## Abstract

Accurate benchmarking of structural variant (SV) detection is essential for advancing the development and application of human whole-genome sequencing (WGS). A fundamental challenge in benchmarking SV detection results is determining whether two SVs represent the same event. Differences in the variation-awareness and strategic implementation of aligners inherently constrain SV detection algorithms that rely on alignment-based approaches. Traditional benchmarking, which primarily focuses on comparing and matching individual variants, makes it difficult to capture the relationships between multiple adjacent variants. We introduced ASVBM, an improved benchmarking framework that introduces the notion of latent positives and leverages a joint analysis and validation strategy based on local variants. This performance improvement arose from the discovery that multiple smaller variants are nearly equivalent to a larger variant. We comprehensively evaluated the performance of six state-of-the-art variant calling pipelines using real WGS datasets. According to multiple matching criteria, ASVBM employs a joint analysis strategy to uncover potential equivalences between the callset and the benchmark set, thereby reducing false mismatches caused by differences in variant representation. ASVBM is available at https://github.com/zhuxiao/asvbm.