A Novel Framework for Predicting Phage-Host Interactions via Host Specificity-Aware Graph Autoencoder.

Area of Science:

• Microbiology and Bioinformatics
• Computational Biology
• Genomics

Background:

• Antibiotic resistance is a growing global health threat, driving the need for alternative treatments like phage therapy.
• Accurate identification of phage-host interactions is crucial for effective phage therapy but remains a significant challenge.
• Existing computational methods often fail to fully utilize phage genomic information or explicitly incorporate host specificity.

Purpose of the Study:

• To develop an efficient computational method for predicting phage-host interactions.
• To address the limitations of existing methods by integrating comprehensive genomic data and host specificity.
• To improve the accuracy and applicability of phage-host interaction prediction for therapeutic purposes.

Main Methods:

• PHISGAE (Phage-Host Interaction prediction using Specificity-aware Graph Autoencoder) was developed.
• Initial phage-phage connections were established using phage genome and protein sequences.
• A refined heterogeneous network was constructed using a K-nearest neighbor strategy.
• A host specificity-aware graph autoencoder was employed to learn phage and bacteria representations for interaction prediction.

Main Results:

• PHISGAE demonstrated superior performance in predicting phage-host interactions compared to state-of-the-art methods.
• Achieved high AUC values of 94.73% at the species level and 96.32% at the genus level.
• Successfully identified potential hosts for previously unseen phages from metagenomic data in case studies.

Conclusions:

• PHISGAE offers an efficient and accurate computational approach for predicting phage-host interactions.
• The method's explicit use of host specificity enhances prediction accuracy.
• PHISGAE has practical implications for identifying novel phage-host pairs in real-world metagenomic applications and advancing phage therapy.