PlasmidHostFinder: Prediction of Plasmid Hosts Using Random Forest

Area of Science:

• Genomics
• Machine Learning
• Antimicrobial Resistance

Background:

• Plasmids are key drivers of antimicrobial resistance (AMR) spread among bacteria.
• Understanding plasmid host range is vital for AMR surveillance and control.
• Current homology-based methods struggle with plasmid diversity and plasticity.

Purpose of the Study:

• To develop an automated method for predicting plasmid host ranges.
• To improve the accuracy of identifying potential bacterial hosts for plasmids.

Main Methods:

• Utilized machine learning, specifically random forests, for host range prediction.
• Trained models on a dataset of 8,519 plasmids from 359 bacterial species.
• Evaluated model performance at species and order taxonomic levels.

Main Results:

• Random forest models achieved high predictive accuracy.
• Matthews correlation coefficients of 0.662 (species) and 0.867 (order) were obtained.
• The developed model accurately distinguishes between plasmid hosts despite plasmid diversity.

Conclusions:

• Machine learning offers a powerful approach to predict plasmid host ranges.
• This tool can enhance AMR surveillance and understanding of gene dissemination.
• An online tool is available for public use in predicting plasmid host ranges.