Graph analysis of fragmented long-read bacterial genome assemblies

Abstract

Motivation Long-read genome assembly tools are expected to reconstruct bacterial genomes nearly perfectly; however, they still produce fragmented assemblies in some cases. It would be beneficial to understand whether these cases are intrinsically impossible to resolve, or if assemblers are at fault, implying that genomes could be refined or even finished with little to no additional experimental cost. Results We propose a set of computational techniques to assist inspection of fragmented bacterial genome assemblies, through careful analysis of assembly graphs. By finding paths of overlapping raw reads between pairs of contigs, we recover potential short-range connections between contigs that were lost during the assembly process. We show that our procedure recovers 45% of missing contig adjacencies in fragmented Canu assemblies, on samples from the NCTC bacterial sequencing project. We also observe that a simple procedure based on enumerating weighted Hamiltonian cycles can suggest likely contig orderings. In our tests, the correct contig order is ranked first in half of the cases and within the top-three predictions in nearly all evaluated cases, providing a direction for finishing fragmented long-read assemblies. Availability and implementation https://gitlab.inria.fr/pmarijon/knot . Supplementary information Supplementary data are available at Bioinformatics online.

Chromosomal rearrangements with stable repertoires of genes and transposable elements in an invasive forest-pathogenic fungus

“…For HybriSPAdes, an optimized k-mer size has to be determined to close gaps, depending on the read length and depth. We tested six different K-mer values (33, 55, 77, 99, 111 and 127) and used Bandage (https://github.com/rrwick/Bandage/wiki/Effect-of-kmer-size) to identify the best k-mer value (Marijon et al, 2019). HybridSPAdes was ran on R1 and R2 paired-end reads corrected with the short read error correction step (BayesHammer; Nikolenko et al, 2013), the accurate mode on, and the Mismatch correction step.…”

“…For HybriSPAdes, an optimized k-mer size has to be determined to close gaps, depending on the read length and depth. We tested six different K-mer values (33, 55, 77, 99, 111 and 127) and used Bandage (https://github.com/rrwick/Bandage/wiki/Effect-of-kmer-size) to identify the best k-mer value (Marijon et al, 2019). We choose the k-mer size producing the assembly graph with the lowest complexity (lowest number of nodes and edges and highest nodes N50), while checking that the graph was not broken into too much disconnected parts as suggested on the Bandage user’s manual (https://github.com/rrwick/Bandage/wiki/Effect-of-kmer-size).…”

Sequencing Technologies and Analyses: Where Have We Been and Where Are We Going?

“…They demonstrate that the method is able to error correct both long and ultra-long sequence reads and is highly scalable, as it is the only method that is able to scale to a human dataset containing ultra-long reads. Marijon et al. (2019) describe a method to analyze assembly graphs produced from long reads to recover contigs that were lost during the assembly process.…”

mosaicFlye: Resolving long mosaic repeats using long error-prone reads

“…Although long reads greatly improved the contiguity of genome assemblies, resolving long repeats remains a challenging task. For example, the state-of-the-art long-read assemblers fail to fully assemble ~50% of bacterial genomes from the NCTC 3000 project aimed at sequencing 3000 bacterial genomes from England's National Collection of Type Cultures (Kamath et al, 2017, Marijon et al, 2019. Additionally, the base-pair accuracy of the long-read assemblies in the repeated regions is reduced (as compared to unique regions) since it is often unclear how to align reads to various repeat copies even if the repeat itself was bridged by some but not all reads.…”