Graph embeddings on gene ontology annotations for protein-protein interaction prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Protein-protein interaction (PPI) prediction is crucial for understanding biological functions, gene-disease, and disease-drug associations.
• Existing PPI prediction methods often overlook missing and spurious interactions within PPI networks.
• Addressing these limitations is vital for accurate biological network analysis.

Purpose of the Study:

• To develop a method for predicting missing and spurious protein-protein interactions (PPIs).
• To leverage graph embeddings for learning vector representations from Gene Ontology Annotation (GOA) graphs.
• To enhance the accuracy of PPI network analysis by accounting for network imperfections.

Main Methods:

• Constructed Gene Ontology Annotation (GOA) graphs incorporating term-term relations and term-protein annotations.
• Employed graph embedding techniques to learn vector representations from the GOA graphs.
• Utilized these learned embeddings to perform missing and spurious PPI prediction tasks.
• Preserved both local and global structural information within the GOA graph.

Main Results:

• The proposed graph embedding method demonstrated superior performance compared to information content (IC)-based and word embedding-based methods.
• Experiments conducted on three PPI datasets from the STRING database validated the method's effectiveness.
• The approach successfully identified missing and spurious interactions in PPI networks.

Conclusions:

• Graph embeddings provide an effective means to learn vector representations from GOA graphs for PPI prediction.
• The method successfully addresses the challenges of missing and spurious interactions in PPI networks.
• This approach enhances the reliability and accuracy of bioinformatics predictions based on PPI data.