Taxonomy-aware, sequence similarity ranking reliably predicts phage-host relationships

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Characterizing phage-host interactions is crucial for understanding microbial ecology and developing phage therapeutics.
• Current experimental methods are slow, low-throughput, and challenging for difficult-to-maintain organisms.
• Existing in silico methods, like BLAST, have limitations in accuracy and detecting divergent relationships.

Purpose of the Study:

• To develop an improved computational tool for predicting prokaryotic hosts of bacteriophages.
• To enhance the accuracy and reliability of phage-host interaction predictions.

Main Methods:

• Developed Phirbo, an extension of BLAST that incorporates sequence similarity and bacterial relatedness.
• Evaluated Phirbo on benchmark datasets of known virus-host pairs.
• Compared Phirbo's performance against alignment-based, alignment-free, and machine-learning host prediction tools.

Main Results:

• Phirbo significantly improved precision and recall by 11-40 percentage points over state-of-the-art tools.
• Achieved a high discriminatory power (AUC = 0.95), outperforming other methods by at least 10 percentage points.
• Demonstrated mean host prediction accuracy of 57% at the genus level and 68% at the family level, with robust performance even with partial viral genome data.

Conclusions:

• Phirbo is a simple, effective, and accurate tool for predicting phage-host relationships.
• The tool enhances the prediction of phage-host interactions, aiding in phage discovery and ecological studies.