A Novel GAER-GMM Framework for Overlapping Protein Complex Detection in Protein Interaction Networks

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Protein complexes are essential for cellular functions.
• Computational methods for detecting protein complexes are advancing.
• Existing methods have limitations in accuracy and generalizability due to overreliance on topology and lack of biological information integration.

Purpose of the Study:

• To develop a novel ensemble learning algorithm for identifying protein complexes in protein interaction networks (PINs).
• To address limitations of existing methods by integrating deep learning and biological features.
• To improve the accuracy and generalizability of protein complex detection.

Main Methods:

• Developed an ensemble learning algorithm integrating a graph autoencoder with a Gaussian Mixture Model (GMM).
• Incorporated protein-related biological features using a specialized feature construction method.
• Applied the algorithm to large-scale datasets (Krogan, DIP, MIPS) and drug target networks.

Main Results:

• The proposed method demonstrated strong performance on multiple large-scale datasets.
• Achieved robust results on drug target networks derived from network pharmacology data.
• Effectively addressed limitations of existing protein complex identification approaches.

Conclusions:

• The novel ensemble learning algorithm offers a promising approach for protein complex identification.
• The integration of graph autoencoders, GMM, and biological features enhances detection accuracy.
• The method shows potential for application in diverse biological network analyses.