Predicting protein complexes in protein interaction networks using Mapper and graph convolution networks

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Protein complexes are crucial for cellular functions and disease mechanisms.
• High-throughput experiments generate large protein-protein interaction datasets.
• Existing computational methods often rely on static protein interaction networks.

Purpose of the Study:

• To develop an advanced computational method for predicting protein complexes.
• To leverage dynamic biological data for more accurate complex identification.
• To improve understanding of cellular processes and disease pathologies.

Main Methods:

• MComplex utilizes time-series gene expression and protein interaction data.
• A temporal network is generated and processed by a generative adversarial network (GAN) with a graph convolutional network (GCN) generator.
• Embeddings are analyzed using a modified graph-based Mapper algorithm for complex prediction.

Main Results:

• MComplex demonstrates superior performance compared to existing methods.
• The method achieves high scores in recall and maximum matching ratio.
• A composite score confirms MComplex's effectiveness in aggregated evaluation measures.

Conclusions:

• MComplex offers a robust and accurate approach for protein complex prediction.
• The integration of dynamic data enhances the identification of protein complexes.
• This method has potential applications in disease mechanism studies and therapeutic development.