An End-to-End Knowledge Graph Fused Graph Neural Network for Accurate Protein-Protein Interactions Prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Artificial Intelligence in Life Sciences

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular functions, disease pathology, and drug discovery.
• Existing artificial intelligence (AI) methods for PPI prediction often suffer from fragmented data handling and suboptimal feature extraction.
• Limitations in non-end-to-end learning frameworks hinder comprehensive analysis of complex biological networks.

Purpose of the Study:

• To develop a novel end-to-end learning model for accurate and comprehensive protein-protein interaction (PPI) prediction.
• To address the limitations of current AI approaches by integrating diverse biological data through a unified framework.
• To improve the prediction accuracy of PPIs by optimizing feature extraction and fusion processes.

Main Methods:

• Construction of a Protein Associated Network (PAN) integrating proteins, drugs, diseases, RNA, and protein structures.
• Application of Graph Neural Networks (GNNs) for extracting topological and semantic features from the PAN and PPI networks.
• Utilization of a multi-layer perceptron for end-to-end feature fusion and PPI prediction.

Main Results:

• The proposed Knowledge Graph Fused Graph Neural Network (KGF-GNN) model demonstrates high accuracy in PPI prediction.
• KGF-GNN significantly outperforms existing state-of-the-art models on real-world PPI datasets.
• The end-to-end learning framework ensures optimized feature extraction and fusion for enhanced prediction.

Conclusions:

• The KGF-GNN model offers a more precise approach to predicting protein-protein interactions.
• This advancement has profound implications for biological research, disease mechanism understanding, and therapeutic development.
• The study highlights the potential of integrated AI approaches in bioinformatics for advancing life sciences.