Benchmarking reveals superiority of deep learning variant callers on bacterial nanopore sequence data

• DOI: 10.7554/elife.98300.2
• Publication Date: 2024-09-05T15:22:44Z
• Authors: Michael B Hall; Ryan R Wick; Louise M Judd; An NT Nguyen; Eike J Steinig; Ouli Xie; Mark R Davies; Torsten Seemann; Timothy P Stinear; Lachlan JM Coin
• Keywords: Benchmarking; Nanopore; Sequence (biology)

Variant calling is fundamental in bacterial genomics, underpinning the identification of disease transmission clusters, construction phylogenetic trees, and antimicrobial resistance prediction. This study presents a comprehensive benchmarking SNP indel variant accuracy across 14 diverse species using Oxford Nanopore Technologies (ONT) Illumina sequencing. We generate gold standard reference genomes project variations from closely-related strains onto them, creating biologically realistic distributions SNPs indels.Our results demonstrate that ONT calls deep learning-based tools delivered higher than traditional methods Illumina, with Clair3 providing most accurate overall. investigate causes missed false calls, highlighting limitations inherent short reads discover ONT’s homopolymer-induced errors are absent high-accuracy basecalling models calls. Furthermore, our findings on impact read depth offer valuable insights for sequencing projects limited resources, showing 10x sufficient to achieve match or exceed Illumina.In conclusion, research highlights superior learning detection sequencing, challenging primacy short-read The reduction systematic ability attain high at lower depths enhance viability widespread use clinical public health genomics.