Tree-Guided Graph Neural Networks with Multilevel Optimization for Protein-Protein Interaction Prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Network Science

Background:

• Predicting human-virus protein-protein interactions (PPI) is vital for understanding viral infections and disease progression.
• Graph neural networks (GNNs) are effective for PPI prediction but struggle with noisy data that disrupts topological information.
• Existing GNNs often fail to capture both global and local patterns crucial for accurate interaction prediction.

Purpose of the Study:

• To develop an advanced GNN model, Tree-Guided Graph Neural Networks with Multi-Level Optimization (TGGNN), for improved human-virus PPI prediction.
• To address the limitations of current GNNs in handling topological disturbances and capturing fine-grained interaction details.
• To enhance the applicability of GNNs in the complex domain of biomedical research and disease mechanism discovery.

Main Methods:

• Proposed TGGNN, inspired by tree nutrient distribution, employing a hierarchically decoupled mechanism to decompose graphs into trunk and leaf subgraphs.
• Developed multilevel semantic subgraphs (trunk graph for core topology, leaf graph for fine-grained patterns) for information aggregation.
• Implemented an attention-based routing mechanism for cross-level semantic fusion and consistent information integration.

Main Results:

• TGGNN demonstrated superior prediction performance compared to state-of-the-art methods across four benchmark datasets.
• The model effectively captured both core topological structures and intricate interaction patterns.
• Case studies confirmed TGGNN's accuracy in identifying specific human-viral PPIs.

Conclusions:

• TGGNN offers a robust and effective approach for predicting human-virus PPIs, outperforming existing methods.
• The hierarchical decomposition and cross-level fusion strategy significantly enhances GNN performance in noisy biological networks.
• TGGNN holds practical utility for advancing biomedical research and identifying potential therapeutic targets.