Enhancing the functional content of eukaryotic protein interaction networks

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• Protein interaction networks are crucial for understanding biological systems.
• Data quality issues like noise and incompleteness hinder network analysis.
• Existing common neighborhood similarity (CNS) measures lack comparative efficacy studies for interaction networks.

Purpose of the Study:

• To evaluate and compare the effectiveness of various common neighborhood similarity (CNS) measures for improving protein interaction network analysis.
• To identify CNS measures that enhance the accuracy of protein function predictions.
• To understand how CNS measures address noise and incompleteness in biological networks.

Main Methods:

• Applied graph transformation to create networks based on different CNS measures.
• Evaluated CNS measure performance by comparing protein function prediction accuracy on transformed versus original networks.
• Utilized large human and fly protein interaction datasets and over 100 Gene Ontology (GO) terms.

Main Results:

• Several transformed networks yielded more accurate protein function predictions than the original network.
• Continuous CNS measures, specifically HC.cont, demonstrated superior performance, especially on large networks.
• CNS measures were found to effectively prune noisy edges and enhance functional coherence.

Conclusions:

• Common neighborhood similarity (CNS) measures can significantly improve the analysis of protein interaction networks.
• Continuous CNS measures offer a robust approach to enhance data quality and prediction accuracy.
• The findings provide a framework for selecting effective CNS measures in bioinformatics research.