Abstract:Abstract Nanopore sequencing has been widely used for the reconstruction of microbial genomes. Owing to higher error rates, errors on the genome are corrected via neural networks trained by Nanopore reads. However, the systematic errors usually remain uncorrected. This paper designs a model that is trained by homologous sequences for the correction of Nanopore systematic errors. The developed program, Homopolish, outperforms Medaka and HELEN in bacteria, viruses, fungi, and metagenomic datasets. When combined with Medaka/HELEN, the genome quality can exceed Q50 on R9.4 flow cells. We show that Nanopore-only sequencing can produce high-quality microbial genomes sufficient for downstream analysis.

Homopolish: a method for the removal of systematic errors in nanopore sequencing by homologous polishing