Multiscale and global-local U-Net for protein-protein interaction site prediction

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Accurate prediction of protein-protein interaction sites (PPIS) is crucial for understanding cellular processes and developing new therapies.
• Existing computational methods often struggle to integrate diverse features, limiting their predictive power.

Purpose of the Study:

• To develop a novel computational architecture, MGU-PPIS, for enhanced protein-protein interaction site prediction.
• To address the limitation of integrating multiscale features capturing both global context and local interactions.

Main Methods:

• Developed the Multiscale and Global-Local U-Net for Protein-Protein Interaction Site Prediction (MGU-PPIS) model.
• Utilized a U-Net framework with multi-level pooling for multiscale feature extraction.
• Integrated Transformer networks, Graph Convolutional Networks (GCNs), and Graph Attention Networks (GATs) within each scale.
• Represented proteins as graphs with residues as nodes and spatial relationships as edges, employing Laplacian positional encoding.
• Implemented a two-stage U-Net architecture for iterative feature refinement.

Main Results:

• MGU-PPIS demonstrated significantly superior predictive accuracy compared to state-of-the-art methods.
• The model effectively captures both global protein structural characteristics and local interaction motifs.
• Experimental validation confirmed the efficacy of the multiscale and global-local feature integration.

Conclusions:

• MGU-PPIS offers a powerful new computational strategy for accurate PPIS prediction.
• The findings advance protein functional analysis and structure-based drug design.
• The proposed architecture provides a foundation for future developments in computational structural biology.