PanSVmerger: a flexible pipeline for merging multiallelic structural variants in pangenome graphs

SummaryPangenome graphs capture extensive genetic diversity but introduce analytical challenges due to the redundant representation of structural variations (SVs). While existing tools effectively address cross-sample redundancy or cross-locus redundancy, none specifically target the intra-locus allelic redundancy inherent to pangenome graphs. Here, we present PanSVmerger, an open-source tool designed to consolidate redundant multiallelic SVs within individual loci using three complementary clustering strategies: adaptive k-mer-based Jaccard distance, global alignment distance via VSEARCH, and length distribution. Validation on HPRC pangenome data demonstrates that PanSVmerger effectively reduces multiallelic complexity (e.g., AC [≥] 3 loci from 62.4% to 4.7% using Strategy A) with a modest trade-off: Recall decreased from 97.13% to 93.58%, while precision improved from 94.95% to 96.56%, yielding an overall F1-score of 95.05%. These results demonstrate that PanSVmerger effectively consolidates redundant allele representations with only a minimal loss of sensitivity, making it well-suited for downstream applications that require clean, non-redundant variants. Availability and implementationPanSVmerger is implemented in Python 3.8+ and freely available under the MIT license at GitHub: https://github.com/tingting100/PanSVmerger. The software requires vcflib, bcftools, and optionally VSEARCH. Comprehensive documentation and tutorials are provided.