DPCMNE: Detecting Protein Complexes From Protein-Protein Interaction Networks Via Multi-Level Network Embedding

Area of Science:

• Cellular Biology
• Bioinformatics
• Computational Biology

Background:

• Protein complexes are essential for cellular functions.
• Accurate detection of protein complexes aids in understanding cellular organization and protein roles.
• Existing computational methods often overlook global network topology, focusing only on local information.

Purpose of the Study:

• To propose a novel computational method, DPCMNE, for detecting protein complexes.
• To integrate both local and global topological information for improved detection accuracy.
• To enhance the understanding of cellular organization and protein functions through precise complex identification.

Main Methods:

• DPCMNE utilizes a multi-level network embedding strategy.
• A hierarchical compression technique recursively reduces the protein-protein interaction (PPI) network.
• Network embedding learns protein representations at various granularities, which are then combined.
• A core-attachment strategy identifies complexes based on learned protein embedding similarities.

Main Results:

• DPCMNE demonstrated superior performance compared to eight other clustering algorithms on two yeast datasets.
• The method achieved higher scores in F1 and F1+Acc metrics.
• Functional enrichment analysis confirmed the greater biological significance of DPCMNE-identified complexes (P-score).

Conclusions:

• DPCMNE effectively captures both local and global network information for protein complex detection.
• The proposed method offers a significant advancement over existing state-of-the-art techniques.
• The identified protein complexes are more biologically relevant, aiding in cellular function research.