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

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Short-read sequencing of bacterial genomes often yields numerous contigs of ambiguous origin (plasmid vs. chromosome).
• Distinguishing plasmid-derived contigs is crucial for understanding the spread of antibiotic resistance and bacterial adaptation.

Purpose of the Study:

• To develop and validate a machine learning model for accurate classification of bacterial DNA contig origin.
• To provide a user-friendly tool for researchers studying bacterial plasmids and antibiotic resistance.

Main Methods:

• Trained machine learning models, specifically Support-Vector Machines (SVM), using pentamer frequencies from complete bacterial genomes (Enterococcus faecium, Klebsiella pneumoniae, Escherichia coli).
• Benchmarked SVM model performance against existing plasmid prediction tools using isolate datasets.
• Validated model scalability on a large collection of E. faecium isolates and demonstrated its application in predicting antibiotic resistance gene locations.

Main Results:

• SVM models achieved high classification accuracy (F1-scores: E. faecium=0.92, K. pneumoniae=0.90, E. coli=0.76), outperforming other methods.
• Successfully predicted the plasmidome for 1644 E. faecium isolates.
• Demonstrated the tool's utility in locating antibiotic resistance genes across the three species.

Conclusions:

• The developed SVM classifiers, available as the 'mlplasmids' R package and GUI, offer a robust solution for bacterial contig origin classification.
• This tool is expected to significantly advance research into plasmid-mediated antibiotic resistance dissemination and bacterial evolution.