Polypolish: Short-read polishing of long-read bacterial genome assemblies.

Wick RR, Holt KE. · 2022

Read Polypolish: Short-read polishing of long-read bacterial genome assemblies. by Wick RR, Holt KE. (2022) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc8812927.txt.

Introduction

Wick and Holt present Polypolish, a tool that fixes errors in bacterial genome assemblies. Long-read sequencing can build a complete bacterial genome. Small errors still remain, most often in runs of one repeated letter. Short reads can correct those errors. Most tools, though, align short reads in ways that fail in repeat regions. Errors there often survive polishing.

Polypolish instead uses all alignments for each read. The authors test it on simulated and real reads. They report that it repairs repeat-region errors other polishers miss. It also almost never adds new errors. The best results came from pairing Polypolish with other polishers. The tool is free software for genome researchers.

Catalog record

CREATORWick RR, Holt KE.
YEAR2022
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC8812927

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Polypolish: Short-read polishing of long-read bacterial genome assemblies By Ryan R. Wick, Kathryn E. Holt, Dina Schneidman-Duhovny Published in: PLoS Computational Biology DOI: 10.1371/journal.pcbi.1009802 ABSTRACT Long-read-only bacterial genome assemblies usually contain residual errors, most commonly homopolymer-length errors. Short-read polishing tools can use short reads to fix these errors, but most rely on short-read alignment which is unreliable in repeat regions. Errors in such regions are therefore challenging to fix and often remain after short-read polishing. Here we introduce Polypolish, a new short-read polisher which uses all-per-read alignments to repair errors in repeat sequences that other polishers cannot. Polypolish performed well in benchmarking tests using both simulated and real reads, and it almost never introduced errors during polishing. The best results were achieved by using Polypolish in combination with other short-read polishers. "This is a PLOS Computational Biology Software paper." INTRODUCTION Long-read-only genome assemblies are inferred using Oxford Nanopore Technologies (ONT) or Pacific Biosciences (PacBio) sequencing reads. For bacterial genomes, reads from these platforms are often longer than the largest repeat in the genome, making complete assemblies (one contig per replicon) possible [1]. However, systematic errors in long reads can lead to hundreds of residual errors in long-read-only assemblies of bacterial genomes, most of which are indels in homopolymer sequences [2, 3]. When these errors occur in protein-coding sequences, they cause frameshifts in the open reading frame, leading to problems with genome annotation and limiting the utility of long-read-only assemblies [4]. Short reads from Illumina platforms do not suffer from the same errors in homopolymer sequences as long reads [5]. Hybrid assembly, using both short and long reads together, can therefore produce sequences which are both complete and highly accurate. Hybrid assembly can be carried out in a short-read-first manner, where long reads are used to scaffold short-read-based contigs into complete genomes, or a long-read-first manner, where short reads are used to correct errors in long-read-based contigs [6]. Given sufficient read depth, long-read-first hybrid assemblies can be more accurate than short-read-first hybrid assemblies, but errors often remain, particularly in repetitive regions of the genome [3]. Long-read-first hybrid assembly commonly contains three stages: long-read assembly, long-read polishing and short-read polishing [7]. All three stages are important when maximising assembly accuracy. The first stage (long-read assembly) aims to create a complete assembly free of any structural errors, i.e. an assembly where the only errors are small in scale. The second stage (long-read polishing) aims to repair as many of the residual small-scale errors as possible using only long reads, e.g. with a platform-specific tool such as Medaka [8]. This study is concerned with the final stage, short-read polishing, which aims to repair any remaining small-scale errors using short reads. There are many short-read polishing tools appropriate for bacterial genomes, including HyPo [9], NextPolish [10], ntEdit [11], Pilon [12], POLCA [13], Racon [14] and wtpoa [15]. Except for ntEdit, which uses a k-mer-based algorithm, each of these tools relies on short-read alignments. The standard approach to short-read alignment involves placing each read in a single location where it best aligns (with ties broken randomly), and this is how popular aligners such as BWA-MEM [16] and Bowtie2 [17] behave with default settings. This method is reliable for non-repetitive parts of the genome, but errors in repeat sequences can cause problems. Specifically, if one instance of a repeat sequence contains an error but another instance does not, reads will preferentially align to the error-free instance, leaving no reads aligned over the error (Fig 1A). Short-read polishing tools which use the alignments may therefore be unable to fix errors in repeats. [Figure: Short-read alignments for assembly polishing.In both examples, the aligned reads (above the line) contain no errors and the assembly sequence (below the line) contains three errors indicated by highlighted columns, one of which is in a repeat sequence. A: standard one-per-read alignments where each read is aligned to a single best location (randomly chosen in a tie). The errors in non-repeat sequences are covered by read alignments, but the error in the repeat has no coverage because reads preferentially aligned to the other instance of the repeat which is error-free. B: all-per-read alignments where each read is aligned to all possible locations. Reads aligned to multiple positions are coloured red. All errors are covered by alignments, including the error in the repeat sequence.] Here, we introduce Polypolish, a new short-read polishing tool which addresses this problem by using a different type of short-read alignment as input. Instead of alignments where each read is placed in a single location, Polypolish is designed to use alignments where each read is aligned to all possible locations (Fig 1B). This ensures that errors in repeats are covered by alignments, allowing Polypolish to fix errors that other short-read polishing tools cannot. RESULTS Implementation Before users run Polypolish, they must align short reads to their long-read assembly using a short-read aligner such as BWA-MEM or Bowtie2. Importantly, the reads must be aligned using the aligner’s all-alignments-per-read option to ensure coverage over genomic repeats (Fig 1B and S1(A) Fig). Polypolish then builds a pileup, where for each position of the assembly sequence, all the read nucleotides associated with that position are collected (S1(B) Fig). ‘Nucleotides’ in the pileup usually consist of a single base, but a read may contribute more than one base per position in the case of an insertion relative to the assembly. When building the pileup, Polypolish trims off two or more bases at the end of each alignment: whatever base is at the end of the read, however many times it occurs, and one more additional base. This trimming ensures that homopolymer indel errors are properly represented in the pileup (S2 Fig). Polypolish then calculates a read depth for each position of the assembly (S1(B) Fig). Reads which align to a single location contribute one unit of depth for each position they cover. Reads which align to multiple locations contribute fractional depth (the reciprocal of their alignment count) to their covered positions; note that this results in approximately the same read depth that would be calculated from the more typical best-match alignment (S3 Fig). Polypolish then assesses each base call in the assembly using the pileup and read depth. For any assembly position at which the pileup contains one and only one valid nucleotide, that valid nucleotide differs from the one currently in the assembly, and all other nucleotides in the pileup are invalid, the assembly base call will be changed to the valid nucleotide (S1(C) Fig). Valid nucleotides are defined as those which have a count greater than the valid threshold: 50% of the read depth at that position, or five, whichever is greater (adjustable by the --fraction_valid and --min_depth options). Invalid nucleotides are defined as those which have a count less than the invalid threshold: 20% of the read depth at that position (adjustable by the --fraction_invalid option). In this manner, Polypolish is a conservative polisher, only making a change where the reads indicate a single unambiguous alternative. In all other cases (e.g. when there are multiple valid nucleotides or there are nucleotides between the two thresholds), the assembly base call is left unchanged. Polypolish is therefore very unlikely to introduce errors during polishing. Simulated-read tests

Read the complete text