A novel link prediction algorithm for protein-protein interaction networks by attributed graph embedding.

Area of Science:

• Computational Biology
• Bioinformatics
• Network Science

Background:

• Protein-protein interactions (PPIs) are crucial for biological processes.
• Computational methods, particularly embedding-based approaches like Deepwalk, are increasingly used for PPI prediction.
• Existing methods often overlook the rich features accompanying proteins in PPI networks.

Purpose of the Study:

• To develop an improved computational method for predicting protein-protein interactions.
• To enhance link prediction accuracy in protein-protein interaction networks by incorporating protein features.
• To present a modified Deepwalk algorithm optimized for attributed networks.

Main Methods:

• Treating PPI prediction as a link prediction problem in attributed networks.
• Developing a modified Deepwalk algorithm incorporating feature selection and weighting.
• Implementing a two-part feature selection process: dimensionality reduction and significance weighting.
• Integrating network structure with selected and weighted protein features for prediction.

Main Results:

• The proposed attributed embedding approach effectively predicts protein-protein interactions.
• The modified Deepwalk algorithm with feature selection outperforms existing network embedding methods.
• Experimental results demonstrate increased prediction accuracy compared to state-of-the-art techniques.

Conclusions:

• The integration of network structure and protein features via feature selection significantly improves PPI prediction.
• The modified Deepwalk algorithm offers a more capable approach for link prediction in biological networks.
• This method provides a valuable tool for understanding complex protein interaction networks.