PhageTB

PhageTB predicts bacteriophage–host interactions to identify suitable phage candidates for phage therapy against pathogenic and antibiotic-resistant bacteria.

Key Features:

• Multilevel Prediction Methodology: Integrates alignment-based and alignment-free features derived from genome sequences of bacteriophages and their hosts to enable predictions across multiple taxonomic levels.
• Alignment-Based Models: Utilizes similarity metrics BLASTPhage (phage-phage), BLASTHost (host-host), and CRISPRPred (phage-CRISPR), achieving accuracies ranging from 42.4% to 80.2% across five taxonomic levels.
• Alignment-Free Models: Constructs machine learning-based alignment-free models that capture sequence patterns without relying on sequence similarity.
• Hybrid Models: Integrates alignment-free model outputs with similarity scores from alignment-based approaches to improve prediction performance, achieving accuracies between 60.6% and 93.5%.
• Ensemble Model: Combines hybrid and alignment-based models to yield accuracies of 67.9%, 80.6%, 85.5%, 90%, and 93.5% at Genus, Family, Order, Class, and Phylum levels, respectively.

Scientific Applications:

• Phage therapy candidate identification: Facilitates identification of potential bacteriophage candidates for treating bacterial infections, including antibiotic-resistant strains.
• Taxonomic-level host prediction: Provides host assignment across Genus, Family, Order, Class, and Phylum levels to inform targeted phage selection.

Methodology:

Integrates alignment-based methods (BLASTPhage, BLASTHost, CRISPRPred) and alignment-free machine learning features derived from genome sequences; hybrid models combine alignment-free outputs with similarity scores and an ensemble integrates hybrid and alignment-based models; trained on 826 phage–host interactions and evaluated on a validation set of 1,201 interactions.