RFPlasmid: predicting plasmid sequences from short-read assembly data using machine learning.

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Antimicrobial-resistance (AMR) genes are frequently located on bacterial plasmids, facilitating their spread.
• Distinguishing between plasmid-borne and chromosomal AMR genes is vital for molecular epidemiology and risk assessment.
• Draft bacterial genome assemblies pose challenges in accurately differentiating plasmid and chromosomal contigs.

Purpose of the Study:

• To develop a novel computational tool, RFPlasmid, for accurate prediction of plasmid versus chromosomal contigs in bacterial genomes.
• To overcome limitations of existing methods that rely on single features and struggle with large, single-copy plasmids.

Main Methods:

• RFPlasmid utilizes a combination of features, including k-mer composition and curated databases of plasmid and chromosomal marker proteins.
• The tool incorporates predictive models for 17 specific bacterial taxa.
• A taxon-agnostic model is available for metagenomic assemblies and unsupported organisms.

Main Results:

• RFPlasmid demonstrates improved accuracy in distinguishing plasmid and chromosomal contigs compared to single-feature methods.
• The tool effectively handles complex genomic data, including large single-copy plasmids.
• Validated models for key bacterial species like *Campylobacter*, *Escherichia coli*, and *Salmonella*.

Conclusions:

• RFPlasmid provides a robust and versatile solution for classifying contigs from draft bacterial genomes.
• Accurate plasmid/chromosome discrimination enhances AMR surveillance, molecular epidemiology, and risk assessment.
• The tool is accessible as a standalone application and through a web interface.