Hidden information revealed by optimal community structure from a protein-complex bipartite network improves protein...

Area of Science:

• Bioinformatics
• Network analysis
• Computational biology

Background:

• Biological networks, like protein-protein interaction (PPI) networks, contain valuable information for predicting protein function.
• While biological networks exhibit community structures, their practical application in function prediction has been limited.
• Existing community-based methods often underperform simple neighbor-based approaches.

Purpose of the Study:

• To develop and evaluate novel methods for protein function prediction on bipartite networks that leverage community structure.
• To compare the performance of community-assisted methods against traditional neighbor-based methods.
• To investigate the impact of community detection strategies on prediction accuracy.

Main Methods:

• Proposed two function prediction approaches for bipartite networks: a screening method and a random forest-based method.
• Incorporated both community structure and neighbor information into the prediction models.
• Evaluated methods on the protein-complex bipartite network of Saccharomyces cerevisiae.

Main Results:

• Community-assisted methods outperformed neighbor-assisted methods for protein function prediction.
• The random forest-based method demonstrated the highest prediction performance.
• Utilizing optimal community structure information was crucial for accurate predictions.
• Projecting bipartite networks into single-mode networks resulted in significant information loss.

Conclusions:

• Community structure information, when properly utilized, enhances protein function prediction accuracy on bipartite networks.
• Random forest models offer a powerful approach for integrating network topology and community information.
• Directly analyzing bipartite networks is superior to projection methods for preserving essential information.