mlplasmids: a user-friendly tool to predict plasmid- and chromosome-derived sequences for single species

Abstract

Assembly of bacterial short-read whole genome sequencing (WGS) data frequently results in hundreds of contigs for which the origin, plasmid or chromosome, is unclear. Long-read sequencing has emerged as a solution to resolve plasmid structures and to obtain complete genomes for most bacterial species. This information can be used to generate and label datasets from short-read based contigs as plasmid-or chromosome-derived. We investigated the use of several popular machine learning methods to classify short-read contigs with known plasmid-or chromosome-origin from Enterococcus faecium , Klebsiella pneumoniae and Escherichia coli using pentamer frequencies. Based on resulting F1-scores we selected support-vector machine (SVM) models as best classifier for all three bacterial species (F1-score E. faecium = 0.94, F1-score K. pneumoniae = 0.90, F1-score E. coli = 0.76) , which outperformed other existing plasmid tools using an independent set of isolates (precision E. faecium = 0.92, precision K. pneumoniae = 0.86, precision E. coli = 0.82). We demonstrated the scalability of our model by accurately predicting the plasmidome of a large collection of 1,644 E. faecium isolates with only short-read WGS available using a standard laptop with a single core. A low number of false positive predicted sequences suggests that the assignment of a particular gene of interest as plasmid-or chromosome-encoded by the models is plausible. The SVM classifiers are publicly available as a new R package called 'mlplasmids' at https://gitlab.com/sirarredondo/mlplasmids under the GNU General Public License v3.0. We additionally developed a graphical-user interface using the Shiny package which can be accessed at https://sarredondo.shinyapps.io/mlplasmids/ . Single genomes can easily be predicted by uploading genome assemblies. We anticipate that this tool may significantly facilitate research on the dissemination of plasmids encoding antibiotic resistance and/or contributing to host adaptation.

Genomes of a major nosocomial pathogenEnterococcus faeciumare shaped by adaptive evolution of the chromosome and plasmidome

“…Detailed description of Illumina and ONT sequencing is available at Supplementary Methods S1-S5 and in Arredondo-Alonso et al (Arredondo-Alonso et al 2018) which includes a full description on ONT selection of E. faecium isolates (n = 62) and consecutive hybrid assembly using Unicycler (Wick et al 2017).…”

“…To determine the plasmidome content of the remaining 1,582 isolates, we used mlplasmids which provides a support-vector machine classifier which was trained and tested using complete genome sequences from E. faecium (Arredondo-Alonso et al 2018). In short, short-read contigs derived from our completed genomes were mapped against the finished chromosomes and plasmids to obtain a short-read contigs dataset labeled either as chromosome-or plasmid-derived.…”

“…We explored the role of plasmids in the dynamics of adaptive evolution of E. faecium by analyzing the total plasmid content (plasmidome) of 62 isolates using a combination of short-read (Illumina) and long-read (Oxford Nanopore Technologies) sequencing. We employed a machine learning approach to predict the plasmidome of 1,582 E. faecium isolates (Arredondo-Alonso et al 2018). We also characterized coevolution of core genomic variation to identify signals of selection and adaptation.…”

Platon: identification and characterization of bacterial plasmid contigs in short-read draft assemblies exploiting protein-sequence-based replicon distribution scores

“…To tackle this issue, many new bioinformatic tools have recently been developed, following different approaches: (i) Recycler and plasmidSPAdes [15,16] exploit coverage variations of sequenced DNA fragments within a genome; (ii) PLACNET investigates paired-end reads linking contig ends [17] ; (iii) PlasmidFinder searches for plasmid specific motifs, i.e. incompatibility groups [18] ; (iv) cBar, PlasFlow and mlPlasmids use machine learning methods to classify k-mer frequencies [19][20][21] ;…”

“…Due to increasing amounts of generated sequence data [24] , there is a rising need for automated high-throughput analysis tools. Unfortunately, not all currently available bioinformatics software tools are suitable for high-throughput analysis, let alone the technical integration into larger analysis pipelines [25][26][27] due to interactive designs or web-based implementations [17,18,21,28] . Taxon-specific database designs also pose additional barriers as users might not have sufficient computational resources or bioinformatics support to build customized or large multi-taxon databases [20,22] .…”