Evaluation of hybrid and non-hybrid methods for de novo assembly of nanopore reads

Abstract

Recent emergence of nanopore sequencing technology set a challenge for the established assembly methods not optimized for the combination of read lengths and high error rates of nanopore reads. In this work we assessed how existing de novo assembly methods perform on these reads. We benchmarked three non-hybrid (in terms of both error correction and scaffolding) assembly pipelines as well as two hybrid assemblers which use third generation sequencing data to scaffold Illumina assemblies. Tests were performed on several publicly available MinION and Illumina datasets of E. coli K-12, using several sequencing coverages of nanopore data (20x, 30x, 40x and 50x). We attempted to assess the quality of assembly at each of these coverages, to estimate the requirements for closed bacterial genome assembly. Results show that hybrid methods are highly dependent on the quality of NGS data, but much less on the quality and coverage of nanopore data and perform relatively well on lower nanopore coverages. Furthermore, when coverage is above 40x, all nonhybrid methods correctly assemble the E. coli genome, even a non-hybrid method tailored for Pacific Bioscience reads. While it requires higher coverage compared to a method designed particularly for nanopore reads, its running time is significantly lower.

Canu: scalable and accurate long-read assembly via adaptivek-mer weighting and repeat separation

“…Table 2 shows Canu assemblies of seven recent 2D Nanopore sequencing runs (http://lab.loman.net/2015/09/24/ first-sqk-map-006-experiment/ and http://lab.loman.net/2016/ 07/30/nanopore-r9-data-release; Loman et al 2015). Consistent with independent evaluations (Judge et al 2016;Sovic et al 2016), Canu produces highly continuous assemblies from Nanopore data alone, and the continuity of Canu assemblies was equal to or better than all assemblers tested. Miniasm was again extremely fast and produced structurally correct and continuous assemblies (Supplemental Note 10, Supplemental Tables S12-S14, Supplemental Figs.…”

“…5A). To date, all assembly evaluations have focused on the more accurate 2D sequences (Loman et al 2015;Judge et al 2016;Sovic et al 2016). While more accurate, the library preparation for 2D sequencing is more complex, reduces the effective throughput of the instrument (each molecule must be read twice), and currently produces shorter sequences.…”

deSALT: fast and accurate long transcriptomic read alignment with de Bruijn graph-based index

“…Although previous studies [36] indicate that databased models more coincide with the characteristics of real sequencing, we considered that real ONT datasets could also have various characteristics [35] and it is hard to build a number of models for benchmark. Thus, we used the two parameter-based models, "ONT 2D reads" and "ONT 1D reads", as a complement, where the 25% and 12% error rates coincide with typical error rates of ONT 2D and 1D reads [18,19]. The parameters and command lines of PBSim and NanoSim are in Supplementary Notes.…”

“…Furthermore, there is less systematic bias in the sequencing procedure [17], which is also beneficial to gene/transcript expression quantification.Besides their advantages, PacBio and ONT reads have much higher sequencing error rates than that of short reads. For PacBio SMRT sequencing, the sequencing error rate of raw reads ("subreads")is about 10% to 20% [16]; for ONT nanopore sequencing, the sequencing error rates of 1D and 2D (also known as 1D 2 ) reads are about 25% and 12% [18,19], respectively. PacBio SMRT platforms can produce reads of inserts (ROIs) by sequencing circular fragments multiple times to largely reduce sequencing errors.…”

“…is about 10% to 20% [16]; for ONT nanopore sequencing, the sequencing error rates of 1D and 2D (also known as 1D 2 ) reads are about 25% and 12% [18,19], respectively. PacBio SMRT platforms can produce reads of inserts (ROIs) by sequencing circular fragments multiple times to largely reduce sequencing errors.…”