A machine learning approach to predict strain-specific phage-host interactions.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Antimicrobial resistance (AMR) necessitates novel therapeutic strategies.
• Bacteriophages (phages) offer a promising biological control alternative.
• Predicting phage-bacteria interactions is essential for effective phage therapy.

Purpose of the Study:

• To develop machine learning (ML) models for predicting phage host range.
• To evaluate the utility of phage-bacteria protein-protein interactions (PPI) as features for ML models.
• To identify sensitive bacterial strains for phage therapy.

Main Methods:

• Developed ML models using phage-bacteria PPI data from databases.
• Trained models with experimental host-range data for Salmonella enterica and Escherichia coli phages.
• Utilized sequencing data and PPI information as input features.

Main Results:

• Prediction model accuracy varied by bacteriophage, ranging from 78-92% for Salmonella and 84-94% for Escherichia.
• Achieved a high accuracy of 94% for predicting the host range of E. coli phage CBDS-07.
• Demonstrated the effectiveness of PPI data in ML models for predicting phage-bacteria interactions.

Conclusions:

• ML models incorporating PPI data are effective for predicting phage-bacteria interactions.
• This approach aids in identifying bacterial strains sensitive to specific phages for therapeutic applications.
• The study highlights the potential of computational methods in advancing phage therapy.