An Ensemble Learning Framework for Detecting Protein Complexes From PPI Networks.

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Protein complex detection is crucial for understanding cellular organization and function.
• Existing computational methods, including unsupervised and supervised learning, have limitations in accuracy and generalization.
• There is a need for robust methods to identify diverse protein complexes within protein-protein interaction (PPI) networks.

Purpose of the Study:

• To develop an advanced computational framework, the Ensemble Learning Framework for Detecting Protein Complexes (ELF-DPC), for accurate protein complex identification.
• To overcome the limitations of existing unsupervised and supervised learning methods in detecting protein complexes with varied topological structures.
• To provide a method that integrates multiple data types and learning strategies for enhanced performance.

Main Methods:

• Constructing a weighted PPI network by integrating topological and biological information.
• Mining protein complex cores using a novel strategy.
• Developing an ensemble learning model combining structural modularity and a trained voting regressor.
• Extending identified cores and forming complete protein complexes via a graph heuristic search.

Main Results:

• The proposed ELF-DPC method demonstrated superior performance compared to twelve state-of-the-art approaches.
• Functional enrichment analysis confirmed that ELF-DPC successfully identifies biologically meaningful protein complexes.
• The framework effectively integrates diverse information for more accurate complex detection.

Conclusions:

• ELF-DPC offers a significant advancement in computational protein complex detection.
• The ensemble learning approach enhances the accuracy and reliability of identifying protein complexes.
• The developed framework provides a valuable tool for biological research, with code and datasets publicly available.