Prokaryotic virus host prediction with graph contrastive augmentaion.

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Bacteriophages (prokaryotic viruses) are vital in microbial community regulation and potential phage therapy agents.
• Accurate prediction of phage-host interactions is critical for understanding viral dynamics and impacts on bacterial populations.
• Existing computational models face limitations due to vast unknown interactions and limited training data diversity.

Purpose of the Study:

• To introduce a novel model, PHPGCA, for enhanced prokaryotic virus host prediction.
• To address the constraints of existing models in handling limited training data and numerous unknown interactions.

Main Methods:

• Constructed a heterogeneous graph integrating virus-virus protein similarity and virus-host DNA sequence similarity.
• Employed LGCN (a graph embedding technique) as the backbone encoder for node representation learning.
• Applied graph contrastive learning for node representation augmentation without requiring additional labels.

Main Results:

• The PHPGCA model demonstrates improved accuracy in predicting phage-host interactions.
• Successfully predicted the host range of multi-species phages in two case studies.
• The approach effectively augments node representations using unlabeled data.

Conclusions:

• PHPGCA offers a robust solution for prokaryotic virus host prediction, overcoming limitations of previous methods.
• The model aids in understanding phage ecology and evolution through accurate host range prediction.
• This work contributes to the advancement of phage therapy applications by improving interaction prediction accuracy.