Double-layer clustering method to predict protein complexes based on power-law distribution and protein...

Area of Science:

• Computational Biology
• Network Science
• Systems Biology

Background:

• Protein complexes are crucial for cellular functions, and identifying them from protein-protein interaction networks (PINs) is a key challenge.
• Existing methods often assume uniform connectivity within complexes, overlooking the diverse topological properties of proteins in biological networks.

Purpose of the Study:

• To develop a new algorithm, PLCluster, for more accurate protein complex identification from PINs.
• To address the limitation of existing methods by considering proteins with different topological characteristics.

Main Methods:

• Proposed a Dense-Spread Centrality method to calculate node centrality scores, which follow a power-law distribution.
• Developed a double-layer clustering approach (PLCluster) that categorizes nodes based on centrality scores and applies distinct detection strategies.
• Implemented a filtering step to remove predicted complexes with proteins in non-matching subcellular compartments.

Main Results:

• PLCluster demonstrated superior performance compared to nine existing methods on a yeast PIN dataset.
• Achieved significant improvements in identifying known complexes, sensitivity, specificity, and f-measure.
• Showed a higher number of perfect matches in complex identification.

Conclusions:

• PLCluster offers a more effective approach to protein complex detection by accounting for varied protein connectivity.
• The method enhances the understanding of protein functions and cellular activities through improved complex identification.
• The integration of centrality distribution and subcellular localization provides a robust framework for network-based biological discovery.