Relational subgraphs fused with complete subgraphs based on the knowledge graph for mining protein complexes

Area of Science:

• Plant molecular biology
• Systems biology
• Bioinformatics

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions and network structures.
• Interpretable domain knowledge discovery is essential for understanding complex PPIs.
• Identifying protein complexes aids in elucidating downstream regulatory genes.

Purpose of the Study:

• To construct a knowledge graph for interacting proteins in Arabidopsis thaliana.
• To develop a relational subgraph-driven model for PPI prediction.
• To discover protein complex structures using mined subgraphs.

Main Methods:

• Constructed a knowledge graph using UniProt and PlaPPISite data for Arabidopsis thaliana.
• Developed a PPI prediction model based on relational subgraphs.
• Extracted complete subgraphs to identify potential protein complexes.

Main Results:

• The knowledge graph contained 68,713 nodes and 109,496 relationships.
• Predicted 1,232 PPIs, with 682 confirmed against STRING and BioGrid databases.
• Identified 336 protein complexes through complete subgraph mining.

Conclusions:

• The integrated knowledge mining method effectively discovers protein complexes.
• This approach offers a novel strategy for analyzing protein complex structures.
• Facilitates the identification of downstream regulatory genes in plants.