Comparing Long-Read Assemblers to Explore the Potential of a Sustainable Low-Cost, Low-Infrastructure Approach to Sequence Antimicrobial Resistant Bacteria With Oxford Nanopore Sequencing

Abstract

Long-read sequencing (LRS) can resolve repetitive regions, a limitation of short read (SR) data. Reduced cost and instrument size has led to a steady increase in LRS across diagnostics and research. Here, we re-basecalled FAST5 data sequenced between 2018 and 2021 and analyzed the data in relation to gDNA across a large dataset (n = 200) spanning a wide GC content (25–67%). We examined whether re-basecalled data would improve the hybrid assembly, and, for a smaller cohort, compared long read (LR) assemblies in the context of antimicrobial resistance (AMR) genes and mobile genetic elements. We included a cost analysis when comparing SR and LR instruments. We compared the R9 and R10 chemistries and reported not only a larger yield but increased read quality with R9 flow cells. There were often discrepancies with ARG presence/absence and/or variant detection in LR assemblies. Flye-based assemblies were generally efficient at detecting the presence of ARG on both the chromosome and plasmids. Raven performed more quickly but inconsistently recovered small plasmids, notably a ∼15-kb Col-like plasmid harboring blaKPC. Canu assemblies were the most fragmented, with genome sizes larger than expected. LR assemblies failed to consistently determine multiple copies of the same ARG as identified by the Unicycler reference. Even with improvements to ONT chemistry and basecalling, long-read assemblies can lead to misinterpretation of data. If LR data are currently being relied upon, it is necessary to perform multiple assemblies, although this is resource (computing) intensive and not yet readily available/useable.

Long read genome assemblers struggle with small plasmids

“…By contrast, Canu recovered 96 % of plasmids in replicate assemblies, followed by Flye-raw (91 %), Flye-meta (90 %), Flye-hq (88 %), Miniasm (85 %), and Raven (75 %) (; ). These findings support previous benchmarking studies which found that Canu and Flye perform best among long read assemblers, in terms of plasmid recovery [6, 11] but indicate that even with the Flye 2.9 update, continued improvements to both assemblers are needed to ensure that all plasmid sequences are recovered.…”

“…Assemblies generated from Illumina short-read sequences have a low error rate but are often highly fragmented, thus making it difficult to confidently differentiate plasmidic sequences from chromosomal sequences. By contrast, long-read sequencing technologies, Oxford Nanopore Technologies (ONT) and PacBio, often produce assemblies that are structurally complete but sometimes miss plasmids due to biases introduced during library preparation [4,5]; and/or issues with long-read genome assemblers [6][7][8]. To overcome these issues, a hybrid approach can be employed, which leverages the strengths of both long-and short-read sequencing technologies, to produce assemblies that are highly accurate and structurally complete [4,9].…”

Nanopore-only assemblies for genomic surveillance of the global priority drug-resistant pathogen, Klebsiella pneumoniae

“…For a 5 Mb genome, such as K. pneumoniae, we previously reported roughly 3000 errors (corresponding to at least 337 substitutions per genome assembly) [38]; therefore, ONTonly assemblies are prone to erroneous SNP calls, which can potentially hamper outbreak investigations [39]. The performance of ONT-only assemblies in identifying AMR and virulence traits against the gold-standard (hybrid assemblies) is unclear but is gaining attention [40][41][42], and is pertinent to inform the research community working with long-read data on what levels of accuracy to expect with current basecalling algorithms.…”

MAECI: A pipeline for generating consensus sequence with nanopore sequencing long-read assembly and error correction

“…Many software or algorithm have been developed for bacterial genome assembly, such as Canu [4], FlyE [5], and Wtdbg2 [6]. They have relative advantages and disadvantages as well as varying performance and assembly outcomes, but in terms of overall performance, FlyE and Raven [7] stands out as the best bacterial genome assembler [8][9][10]. Nanopore sequencing data are characterized by the presence of indels, non-random systematic errors [11] and the occurrence of assembly errors spanning hundreds of bases [8], which may lead to inaccurate or incomplete assemblies.…”