NextPolish2: A Repeat-aware Polishing Tool for Genomes Assembled Using HiFi Long Reads

Abstract

The high-fidelity (HiFi) long-read sequencing technology developed by PacBio has greatly improved the base-level accuracy of genome assemblies. However, these assemblies still contain base-level errors, particularly within the error-prone regions of HiFi long reads. Existing genome polishing tools usually introduce overcorrections and haplotype switch errors when correcting errors in genomes assembled from HiFi long reads. Here, we describe an upgraded genome polishing tool — NextPolish2, which can fix base errors remaining in those “highly accurate” genomes assembled from HiFi long reads without introducing excessive overcorrections and haplotype switch errors. We believe that NextPolish2 has a great significance to further improve the accuracy of telomere-to-telomere (T2T) genomes. NextPolish2 is freely available at https://github.com/Nextomics/NextPolish2.

1. Introduction

Complete and accurate genomes provide fundamental tools for scientists to capture a full spectrum of genomic variants and use such information to understand the evolutionary basis of various diseases and other biological phenotypes [1]. Hence, a complete and gapless genome, also known as a telomere-to-telomere (T2T) genome, has been emerging as a new hotspot in the field of genomics [2–8]. Typically, we obtain a T2T genome with datasets including both high-accuracy PacBio high-fidelity (HiFi) long reads and Oxford Nanopore Technologies (ONT) error-prone ultra-long reads [2]. Compared to the genomes generated using noisy long reads, genomes obtained using HiFi long reads have considerably higher quality — far fewer errors at the level of single nucleotides and small insertions and deletions (InDels) [9,10]. However, they still contain a handful of assembly errors in chromosomal regions where HiFi long reads stumble as well, such as homopolymer or low-complexity microsatellite regions (Figures S1 and S2). Additionally, a typical gap-filling step is accomplished using ONT ultra-long reads which contain a certain amount of errors that need to be corrected [11]. Hence, the current T2T genomes assembled using the cutting-edge sequencing platforms still require further improvement in terms of consensus accuracy. For example, the Human Genome T2T Consortium has applied multiple tools and extensive manual validation to increase the assembly quality value (QV) from 70.2 to 73.9 for the T2T assembly of a human genome (CHM13) [11].

Error correction for a T2T genome assembly is challenging because (1) complex segmental duplications and large tandem repeats, such as centromeric satellite arrays, could potentially induce overcorrections or false negative corrections; (2) local haplotype needs to be maintained; and (3) there are technology-specific biases for different sequencing platforms [12]. Therefore, although there are many state-of-the-art polishing tools available, such as Pilon [13], Racon [14], and NextPolish [15], these tools were designed for genomes assembled from noisy long reads and mainly aimed at errors at levels of single nucleotides and small InDels. They can hardly handle assembly errors sourced from HiFi long reads, without taking the aforementioned challenges into consideration.