Explainable spatio-temporal graph evolution learning with applications to dynamic brain network analysis during...

Area of Science:

• Graph learning
• Network science
• Computational neuroscience

Background:

• Modeling dynamic interactions in complex networks is essential for understanding their evolution.
• Existing spatio-temporal graph learning methods have limitations in exploiting spatial neighborhoods and capturing temporal dependencies of inter-node relations (INRs).
• Explainability of these models remains an understudied area.

Purpose of the Study:

• To propose an explainable spatio-temporal graph evolution learning (ESTGEL) model.
• To effectively model the dynamic evolution of inter-node relations (INRs) in complex networks.
• To enhance the explainability of network evolution models.

Main Methods:

• Developed an edge attention module to leverage multi-level spatial neighborhoods of INRs.
• Introduced a dynamic relation learning module to capture spatio-temporal dependencies.
• Integrated INRs into node representations for comprehensive network evolution analysis.

Main Results:

• The ESTGEL model was validated on a brain development study dataset.
• Experimental results demonstrated that brain functional networks transition from dispersed to more convergent and modular structures during development.
• Significant changes in dynamic functional connectivity (dFC) were observed, particularly in areas related to emotional control, decision-making, and language processing.

Conclusions:

• The proposed ESTGEL model offers a novel approach for understanding complex network evolution.
• The findings highlight developmental shifts in brain network organization, moving towards increased modularity.
• The study provides insights into the dynamic functional connectivity changes underlying cognitive development.