SAVANA: reliable analysis of somatic structural variants and copy number aberrations in clinical samples using long-read sequencing

Accurate detection of somatic structural variants (SVs) and copy number aberrations (SCNAs) is critical to inform the diagnosis and treatment of human cancers. Here, we describe SAVANA, a computationally efficient algorithm designed for the joint analysis of somatic SVs, SCNAs, tumour purity and ploidy using long-read sequencing data. SAVANA relies on machine learning to distinguish true somatic SVs from artefacts and provide prediction errors for individual SVs. Using high-depth Illumina and nanopore whole-genome sequencing data for 99 human tumours and matched normal samples, we establish best practices for benchmarking SV detection algorithms across the entire genome in an unbiased and data-driven manner using simulated and sequencing replicates of tumour and matched normal samples. SAVANA shows significantly higher sensitivity, and 9- and 59-times higher specificity than the second and third-best performing algorithms, yielding orders of magnitude fewer false positives in comparison to existing long-read sequencing tools across various clonality levels, genomic regions, SV types and SV sizes. In addition, SAVANA harnesses long-range phasing information to detect somatic SVs and SCNAs at single-haplotype resolution. SVs reported by SAVANA are highly consistent with those detected using short-read sequencing, including complex events causing oncogene amplification and tumour suppressor gene inactivation. In summary, SAVANA enables the application of long-read sequencing to detect SVs and SCNAs reliably in clinical samples.