Time-sequential graph adversarial learning for brain modularity community detection.

Area of Science:

• Neuroscience
• Network Science
• Machine Learning

Background:

• Brain community detection is crucial for understanding brain network organization.
• Dynamic changes in brain function and complex community structures present significant challenges.
• Existing methods struggle to capture the temporal dynamics and intricate community architectures.

Purpose of the Study:

• To propose a novel framework for brain community detection that addresses temporal dynamics.
• To characterize the structure of communities within brain networks.
• To enhance the accuracy and robustness of brain community detection.

Main Methods:

• Developed a time-sequential graph neural network (TGAL) with a spatio-temporal attention mechanism for feature extraction.
• Incorporated a modularity loss function to optimize community structure by maximizing modularity.
• Employed an adversarial learning scheme to guide representation learning.

Main Results:

• The TGAL framework demonstrated superior performance in brain community detection on real-world datasets.
• The model effectively captured spatio-temporal dynamics and intricate community structures.
• Experimental results validated the advantage of the proposed framework over existing methods.

Conclusions:

• The TGAL framework offers an effective solution for dynamic brain community detection.
• The proposed method advances the characterization of brain network communities.
• This work provides a powerful tool for neuroscience research and brain data analysis.