Accurately Detecting Protein Complexes by Graph Embedding and Combining Functions with Interactions.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Protein complex identification is vital for understanding cellular mechanisms and drug development.
• Existing computational methods for protein complex detection primarily rely on protein-protein interaction networks (PINs).
• High rates of false positives/negatives in PINs limit the effectiveness of topology-based detection methods.

Purpose of the Study:

• To develop a novel computational approach for enhanced protein complex detection.
• To merge topological information from PINs with functional protein information for improved accuracy.
• To overcome limitations of existing methods that solely use network topology.

Main Methods:

• Proteins are grouped based on functional information (FunCat data).
• Protein-protein similarity matrices are computed using graph embedding on PINs and Gene Ontology (GO) terms.
• An integrated similarity matrix is created by combining topological and functional similarity.
• Proteins within each group are clustered using the integrated similarity matrix, and resulting clusters are mapped to PINs.
• Overlapping subgraphs are merged to identify final protein complexes.

Main Results:

• The proposed method successfully identifies protein complexes by integrating diverse data sources.
• Empirical evaluations on multiple benchmark datasets (Collins, Gavin, Krogan, Wiphi, CYC2008, MIPS) demonstrate superior performance.
• The integrated approach significantly outperforms existing state-of-the-art protein complex detection methods.

Conclusions:

• Combining protein-protein interaction network topology with functional information offers a more robust approach to protein complex detection.
• This method enhances the accuracy and reliability of identifying protein complexes.
• The findings have implications for advancing cellular function studies and facilitating targeted drug design.