Predicting phage-host interaction via hyperbolic Poincaré graph embedding and large-scale protein language technique

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Bacteriophages (phages) show promise for treating antibiotic-resistant bacterial infections.
• Identifying specific phages for targeted therapy is challenging due to diverse host ranges.
• Current computational methods lack accuracy in predicting phage-host interactions across bacterial species.

Purpose of the Study:

• To develop GE-PHI, a novel machine-learning model for accurate prediction of phage-host interactions (PHIs).
• To integrate knowledge graph embedding and protein language models for enhanced PHI prediction.
• To provide a tool for advancing phage therapeutics and diagnostics in microbial engineering.

Main Methods:

• Constructed a phage-host heterogeneous association network (PHAN) including phage-phage and host-host similarity.
• Applied multi-relational Poincaré graph embedding (MuRP) to extract network topological patterns.
• Utilized the ESM-2 protein language model to capture evolutionary information from phage and host proteins.

Main Results:

• GE-PHI achieved a cross-validation area under the curve (AUC) of 0.9453 in silico.
• The model demonstrated robust performance in case studies, validating its predictive accuracy.
• Integrated network topology and protein evolutionary information for superior PHI prediction.

Conclusions:

• GE-PHI offers a significant advancement in predicting phage-host interactions.
• The model supports the development of machine-learning-guided phage therapeutics.
• GE-PHI aids in phage diagnostics and microbial engineering applications.