Joint Identification of Network Hub Nodes by Multivariate Graph Inference

Recent development of neuroimaging technique allows us to investigate the structural and functional connectivity of our brain in vivo. Since hub nodes are often located at the critical regions and exhibit special integrative or control functions in our brain, identification of hubs from network data has attracted much attention in neuroscience. Current state-of-the-art methods usually select the hub nodes one after another based on either the heuristics of connectivity profile at each node or the predefined setting of network modules. Thus, current computational methods have limited power to recognize connector hubs which link multiple modules and thus have higher importance than provincial hubs (centers of module with large connectivity degrees). To address this challenge, we propose a novel multivariate hub identification method to simultaneously estimate the setting of connector hubs towards the optimal scenario where the removal of these identified hubs brings the worst catastrophe to the original network. We have compared our hub identification method with the existing methods on both simulated and real network data. Our proposed method achieves more accurate and replicable result of hub nodes which shows the enhanced statistical power in distinguishing network alterations related to neuro-disorders such as Alzheimer's disease.