Dynamics of functional network organization through graph mixture learning.

Area of Science:

• Network neuroscience
• Computational neuroscience
• Graph theory applications in neuroscience

Background:

• Understanding human brain activity organization at the systems level is a significant challenge.
• Existing methods often rely on structural information or time-windowing, which have limitations.

Purpose of the Study:

• To develop a fully data-driven approach for extracting repeating network patterns from brain activity.
• To infer functional brain networks without relying on structural connectome information.
• To overcome limitations associated with time-windowing functional signals.

Main Methods:

• Utilized a Graph Laplacian Mixture Model (GLMM), a generative model for functional data.
• Employed graph learning to extract patterns from regionally-averaged timecourses by exploiting covariance between brain region activities.
• Applied the GLMM to both task fMRI and resting-state fMRI data.

Main Results:

• The GLMM successfully identified meaningful repeating network patterns from functional brain activity.
• Learned graphs showed high similarity to the structural connectome, despite not using structural information.
• The Default Mode Network (DMN) was consistently captured across different tasks and resting-state conditions.
• Extracted patterns aligned with known brain activation patterns and task timing.

Conclusions:

• The GLMM provides a powerful, data-driven method for inferring functional brain networks.
• This approach effectively identifies key brain networks, including the DMN, without prior structural information.
• The method handles the entire functional signal, avoiding issues related to time-windowing.