MED-PPIS: Multi-order Moments External Graph Attention Network with Dual-Axis Attention for Protein-Protein...

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Accurate prediction of protein-protein interaction (PPI) sites is crucial for understanding cellular mechanisms and advancing genomics.
• Current graph neural networks (GNNs) face limitations in capturing inter-graph correlations and accurately representing feature distributions.

Purpose of the Study:

• To introduce MED-PPIS, a novel framework designed to overcome the limitations of existing GNNs for PPI site prediction.
• To enhance the accuracy and efficiency of predicting protein-protein interactions.

Main Methods:

• MED-PPIS integrates an mLSTM-based matrix memory for long-range dependencies and a multi-order moment GNN for feature distributions.
• A graph external attention mechanism learns universal structural motifs, complemented by a dual-axis attention for multi-scale feature extraction.

Main Results:

• MED-PPIS demonstrated significant improvements over the strongest baseline on the Test_60 dataset.
• Key metrics showed increases: 2.1% in AUPRC, 1.2% in AUROC, and 2.3% in F1-score.

Conclusions:

• MED-PPIS offers a powerful and transparent tool for analyzing protein interaction landscapes.
• The framework facilitates new biological insights and the development of therapeutic strategies through superior predictive accuracy.