Graph neural network integrated with pretrained protein language model for predicting human-virus protein-protein...

Area of Science:

• Computational Biology
• Virology
• Bioinformatics

Background:

• Understanding human-virus protein-protein interactions (PPIs) is crucial for combating viral infections.
• Existing methods struggle with novel viral threats, necessitating advanced predictive tools.

Purpose of the Study:

• To develop DeepGNHV, an advanced framework for predicting human-virus PPIs.
• To enhance the detection of interactions involving emerging viral pathogens.

Main Methods:

• Integrated a pretrained protein language model with AlphaFold2-derived structural features.
• Utilized graph attention networks for end-to-end PPI prediction.
• Validated robustness across perturbations and high-confidence predictions.

Main Results:

• DeepGNHV demonstrated superior predictive performance compared to state-of-the-art methods.
• Achieved strong generalization capabilities for predicting interactions with novel viral proteins.
• Identified specific host factors interacting with high-risk Human Papillomavirus (HPV).

Conclusions:

• DeepGNHV offers a robust and generalizable approach for human-virus PPI prediction.
• The framework aids in understanding viral infection mechanisms and developing targeted interventions.
• Case studies pinpointed critical residues involved in specific PPIs, offering insights into molecular mechanisms.