SVScope improves somatic structural variations detection via graph-genome optimization

Somatic structural variations (SVs) are critical in cancer genomes, yet their detection from long-read sequencing remains challenging due to alignment errors in repetitive regions. We develop SVScope, leveraging full-length reads and local graph-genome optimization with a random forest strategy to improve somatic SV calling. We also provide ScopeVIZ, a companion pipeline for visualizing read clustering at breakpoints. Across seven benchmark cell lines sequenced with ONT and PacBio platforms, as well as simulated datasets, SVScope consistently outperforms state-of-the-art methods, achieving up to 23.64% improvement in F1-score. Using SVScope, we validate 32 somatic SVs, expanding the ground-truth dataset by 47.06%.