mlplasmids

mlplasmids classifies bacterial contigs as plasmid- or chromosome-derived for Enterococcus faecium, Escherichia coli, and Klebsiella pneumoniae using support-vector machine (SVM) binary classifiers trained on labeled short-read contigs derived from long-read–resolved complete genomes.

Key Features:

• Machine Learning Approach: Uses support-vector machine (SVM) binary classifiers trained on pentamer (5-mer) frequency features from labeled short-read contigs.
• Performance Metrics: Reported F1-scores are 0.92 for E. faecium, 0.90 for K. pneumoniae, and 0.76 for E. coli.
• Training Data: Models are trained on short-read contigs labeled using complete genomes resolved by long-read sequencing.
• Scalability: Demonstrated application to 1,644 E. faecium isolates for plasmidome prediction.
• Antibiotic Resistance Gene Localization: Facilitates prediction of antibiotic resistance gene locations as plasmid- or chromosome-derived.
• Benchmarking: Validated and benchmarked against existing plasmid prediction tools.

Scientific Applications:

• Plasmidome Analysis: Predicts plasmid-derived contigs to support pan-plasmidome characterization.
• Source Specificity Studies: Enables analyses of source specificity in pathogenic bacteria such as E. faecium by distinguishing plasmid and chromosomal origins.
• Ecological and Evolutionary Insights: Provides information on plasmid population configurations, ecological constraints, horizontal gene transfer, and niche adaptation.

Methodology:

Classifiers are trained on labeled short-read contigs derived from complete genomes resolved by long-read sequencing using pentamer frequency features; SVM models were selected for classification and performance was validated by benchmarking against other plasmid prediction tools.