Learning Functional Dynamics from a Multilayer Brain Network

Mounting evidence reveals that functional brain network connectivity undergoes dynamic changes over time, with various disease-related alterations implicitly woven into these functional dynamics. However, existing computational approaches, particularly those using the multigraph model, often focus solely on temporal topological connectivity changes across windows, overlooking the underlying causal evolution patterns between windows. To fully capture the intricate functional dynamics of the brain, we propose a spatio-temporal dynamics learning framework that integrates temporal connectivity within windows and causal relationships between them into a unified multilayer network structure. Specifically, our multilayer graph embedding learning on manifolds can effectively preserve the intrinsic evolution pattern of functional dynamics. Experimental results on real-world datasets demonstrate that the incorporation of underlying causal relationships can significantly discover more disease-related functional dynamics.