PTBGRP: predicting phage-bacteria interactions with graph representation learning on microbial heterogeneous...

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Bacteriophages (phages) are crucial for combating bacterial infections, necessitating effective prediction methods.
• Current computational approaches for phage prediction often overlook complex, higher-order biological network patterns.
• Integrating diverse biological information can enhance the accuracy of predicting phage-bacteria interactions.

Purpose of the Study:

• To develop a novel computational model, PTBGRP, for predicting bacteriophage candidates against bacterial hosts.
• To improve the accuracy of phage-bacteria interaction (PBI) prediction by incorporating higher-order network features.
• To provide a freely accessible webserver for the PTBGRP predictor.

Main Methods:

• Constructed a microbial heterogeneous interaction network (MHIN) integrating PBI and bacteria-bacteria interaction data.
• Employed representation learning to extract high-level biological and topological features from the MHIN.
• Utilized a deep neural network classifier to predict unknown PBI pairs based on fused features.

Main Results:

• PTBGRP demonstrated superior performance compared to state-of-the-art methods on the ESKAPE pathogens dataset.
• Case studies on Klebsiella pneumoniae and Staphylococcus aureus validated PTBGRP's predictive accuracy.
• The model's effectiveness is attributed to the comprehensive integration of heterogeneous biological information.

Conclusions:

• The PTBGRP model offers an advanced computational approach for identifying potential phage therapies.
• Integrating higher-order network connectivity significantly enhances the prediction of phage-bacteria interactions.
• PTBGRP provides a valuable tool for researchers in the fight against bacterial pathogens.