Multi-hierarchy Network Configuration Can Predict Brain States and Performance.

The brain is a hierarchical modular organization that varies across functional states. Network configuration can better reveal network organization patterns. However, the multi-hierarchy network configuration remains unknown. Here, we proposed an eigenmodal decomposition approach to detect modules at multi-hierarchy, which can identify higher-layer potential submodules, and is consistent with the brain hierarchical structure. We defined three metrics: node configuration matrix, combinability, and separability. Node configuration matrix represents network configuration changes between layers. Separability reflects network configuration from global to local, whereas combinability shows network configuration from local to global. First, we created a random network to verify the feasibility of the method. Results show that separability of real networks is larger than that of random networks, whereas combinability is smaller than random networks. Then, we analyzed a large data set incorporating fMRI data from resting and seven distinct tasking conditions. Experiment results demonstrates the high similarity in node configuration matrices for different task conditions, whereas the tasking states have less separability and greater combinability between modules compared with the resting state. Furthermore, the ability of brain network configuration can predict brain states and cognition performance. Crucially, derived from tasks are highlighted with greater power than resting, showing that task-induced attributes have a greater ability to reveal individual differences. Together, our study provides novel perspectives for analyzing the organization structure of complex brain networks at multi-hierarchy, gives new insights to further unravel the working mechanisms of the brain, and adds new evidence for tasking states to better characterize and predict behavioral traits.