Supervised maximum-likelihood weighting of composite protein networks for complex prediction

Area of Science:

• Biochemistry and Molecular Biology
• Systems Biology
• Bioinformatics

Background:

• Protein complexes are crucial for cellular functions, necessitating accurate identification for understanding cellular organization and regulation.
• High-throughput protein-protein interaction (PPI) data aids in discovering protein complexes but is hampered by noise, including spurious and missing interactions.
• Existing methods struggle with transient interactions and incomplete complex memberships, leading to inaccurate predictions and missed true complexes.

Purpose of the Study:

• To develop a robust method for discovering protein complexes from noisy protein-protein interaction data.
• To improve the accuracy, precision, and recall of protein complex identification.
• To facilitate the discovery of novel protein complexes and assess their functional significance.

Main Methods:

• Integrated protein-protein interaction (PPI) data with heterogeneous data sources to build a composite protein network.
• Employed a supervised maximum-likelihood approach to weight network edges based on their probability of belonging to a complex.
• Utilized six clustering algorithms and an aggregative clustering strategy for complex discovery in the weighted network.

Main Results:

• The developed method demonstrated improved protein complex discovery in Saccharomyces cerevisiae and Homo sapiens.
• Achieved higher precision and recall compared to previous supervised and unsupervised weighting approaches.
• Generated novel protein complexes with greater functional similarity and provided visualizations for credibility assessment.

Conclusions:

• The integrated approach, combining multiple data sources and supervised learning, effectively creates a weighted composite protein network for novel complex discovery.
• The method significantly outperforms previous approaches in precision, recall, and the quality of novel predictions.
• External evidence supports the validity of newly predicted complexes, aiding in the understanding of yeast and human cellular machinery.