LGS-PPIS: A Local-Global Structural Information Aggregation Framework for Predicting Protein-Protein Interaction...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning in Biology

Background:

• Protein-protein interactions (PPIs) are crucial for biological processes.
• Experimental determination of protein-protein interaction sites (PPIS) is costly and time-consuming.
• Existing deep learning methods for PPIS prediction have limitations in feature extraction.

Purpose of the Study:

• To address limitations in current PPIS prediction methods.
• To develop a novel framework for enhanced PPIS prediction.
• To improve the accuracy and efficiency of identifying protein interaction sites.

Main Methods:

• Proposed a local-global structural information aggregation framework (LGS-PPIS).
• Incorporated an edge-aware graph convolutional network (EA-GCN) for local feature extraction.
• Utilized a self-attention mechanism with initial residual and identity mapping (SA-RIM) for global feature extraction.

Main Results:

• LGS-PPIS outperformed state-of-the-art deep learning methods on three PPIS prediction benchmarks.
• Ablation studies confirmed the benefits of both local (EA-GCN) and global (SA-RIM) feature aggregation.
• Local features from spatially adjacent residues played a more significant role in PPIS prediction.

Conclusions:

• The LGS-PPIS framework effectively integrates local and global structural information for improved PPIS prediction.
• The EA-GCN module is particularly important for capturing essential local features.
• This approach offers a more accurate and efficient alternative for PPIS determination, aiding in the elucidation of biological mechanisms.