Brain Connectivity: Exploring from a High-Level Topological Perspective.

How invariant structural architecture of brain coupling with variant functionality is still unclear in neuroscience. The previous exploration of relationships between large-scale structural and functional brain networks mainly focused on whole or partial statistical correlation, ignoring network context information, such as network topology structure. Here we applied a network representation learning approach to create high-order representations of structural or functional networks while preserving network context information for studying the function-structure coupling of the brain at topological subnetwork levels. We found that the structural and functional network obtained from the network representation learning method was more stable and more tightly coupled than those from the conventional correlation method, primarily distributed in high-order cognitive networks. Application on schizophrenia patients showed decoupling on the default-mode network, dorsal attention network, executive control network, and salience network, as well as the over-coupling on the sensorimotor network, compared with healthy controls. Overall, network representation learning can more effectively capture the higher-order coupling between brain structure and function and provides a good technical means for us to study mental illness.