[A protein complex recognition method based on spatial-temporal graph convolution neural network].

Area of Science:

• Computational biology
• Network science
• Bioinformatics

Background:

• Dynamic protein networks are crucial for understanding cellular processes.
• Identifying protein complexes within these networks is a fundamental challenge.
• Existing methods often struggle with the temporal and structural complexities of dynamic networks.

Purpose of the Study:

• To introduce a new method for mining protein complexes in dynamic protein networks.
• To leverage spatiotemporal convolution neural networks for enhanced network analysis.
• To improve the accuracy and efficiency of protein complex identification.

Main Methods:

• Developed a dynamic protein network model incorporating edge strength, node strength, and edge existence probability.
• Designed two convolution operators using Hilbert-Huang transform, attention mechanisms, and residual connections.
• Constructed a dynamic protein network characteristic map and employed spectral clustering for complex identification.

Main Results:

• Achieved an F-value exceeding 90% on the DIP and MIPS datasets.
• Demonstrated significant improvements in recognition efficiency over four other algorithms (DPCMNE, GE-CFI, VGAE, NOCD).
• The proposed algorithm showed efficiency gains of 34.5%, 28.7%, 25.4%, and 17.6% respectively.

Conclusions:

• Deep learning, specifically spatiotemporal convolution neural networks, offers a powerful approach for analyzing dynamic protein networks.
• The proposed method significantly enhances the efficiency and accuracy of protein complex identification.
• This work contributes to advancing computational methods in systems biology and network analysis.