Evaluation of clustering algorithms for protein complex and protein interaction network assembly

Area of Science:

• Proteomics
• Systems Biology
• Bioinformatics

Background:

• Assembling protein complexes and interaction networks from proteomics data is difficult, especially with limited prior knowledge.
• Evaluating clustering results for protein networks requires robust methodologies.

Purpose of the Study:

• To systematically compare hierarchical and partitioning clustering approaches for analyzing complex human protein interaction networks.
• To assess the utility of different data normalization methods (binary, normalized spectral abundance factors, Z-score) in network assembly.

Main Methods:

• Systematic comparison of multiple hierarchical and partitioning clustering algorithms.
• Utilized a well-characterized human protein interaction dataset centered around AAA+ ATPases Tip49a and Tip49b.
• Evaluated binary data, normalized spectral abundance factors, and Z-score normalization.

Main Results:

• Partitioning approaches effectively identified major network modules.
• Euclidean distance required data transformation to recover all network attachments.
• Pearson correlation combined with hierarchical clustering successfully separated protein complexes and correctly placed their attachments.

Conclusions:

• Different clustering approaches provide distinct, valuable information for assembling complex protein interaction networks.
• Pearson correlation and hierarchical clustering are effective for resolving individual protein complexes within larger networks.