Predicting bacteriophage hosts based on sequences of annotated receptor-binding proteins.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Bacteriophages are promising alternatives to antibiotics for treating bacterial infections.
• Characterizing phage host specificity is currently time-consuming and labor-intensive.
• The ESKAPE group (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacter species), Escherichia coli, Salmonella enterica, and Clostridium difficile are critical pathogens.

Purpose of the Study:

• To develop a machine learning pipeline for predicting bacteriophage hosts based on receptor-binding protein (RBP) sequences.
• To evaluate the model's performance against the Basic Local Alignment Search Tool (BLASTp).
• To identify applications for machine learning in phage therapy and RBP engineering.

Main Methods:

• A machine learning model was trained on annotated RBP sequence data.
• The model was used to predict hosts for bacteria including the ESKAPE group, E. coli, S. enterica, and C. difficile.
• Performance was assessed using Precision-Recall Area Under the Curve (PR-AUC) and compared to BLASTp.

Main Results:

• The best-performing model achieved PR-AUC scores ranging from 73.6% to 93.8%.
• The model's performance was comparable to BLASTp for high sequence similarity.
• The machine learning model outperformed BLASTp when sequence similarity dropped below 75%.

Conclusions:

• Machine learning offers an efficient method for predicting bacteriophage hosts from RBP sequences.
• This approach is particularly valuable for novel sequences with low similarity to existing databases.
• Predicting hosts of metagenomic RBP sequences can enhance phage therapy and bacteriocin development by enabling RBP swapping to tune host range.