Graph embedding-based novel protein interaction prediction via higher-order graph convolutional network.

Area of Science:

• Computational Biology
• Network Science
• Bioinformatics

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions, but current PPI networks are noisy and incomplete.
• Understanding cellular systems requires accurate and comprehensive PPI networks.

Purpose of the Study:

• To develop a computational method for predicting missing protein-protein interactions (PPIs) in noisy networks.
• To improve the accuracy and robustness of protein interaction prediction using network topology alone.

Main Methods:

• Proposed a novel node embedding method combining Graph Convolutional Networks (GCNs) and PageRank to capture higher-order network topology.
• Developed a higher-order GCN variational auto-encoder (HO-VGAE) architecture for joint node representation learning.
• Focused exclusively on network topology, without using protein attributes or external biological features.

Main Results:

• The HO-VGAE method demonstrated competitive performance in PPI prediction.
• The proposed method outperformed existing graph embedding-based link prediction techniques in both accuracy and robustness.
• The approach successfully predicted novel protein interactions based solely on network structure.

Conclusions:

• The novel HO-VGAE method effectively predicts protein-protein interactions using only network topology.
• This approach offers a robust and accurate way to expand the human interactome and identify potential interactions for experimental validation.