BNPower: a power calculation tool for data-driven network analysis for whole-brain connectome data

Abstract

Abstract Network analysis of whole-brain connectome data is widely employed to examine systematic changes in connections among brain areas caused by clinical and experimental conditions. In these analyses, the connectome data, represented as a matrix, are treated as outcomes, while the subject conditions serve as predictors. The objective of network analysis is to identify connectome subnetworks whose edges are associated with the predictors. Data-driven network analysis is a powerful approach that automatically organizes individual predictor-related connections (edges) into subnetworks, rather than relying on pre-specified subnetworks, thereby enabling network-level inference. However, power calculation for data-driven network analysis presents a challenge due to the data-driven nature of subnetwork identification, where nodes, edges, and model parameters cannot be pre-specified before the analysis. Additionally, data-driven network analysis involves multivariate edge variables and may entail multiple subnetworks, necessitating the correction for multiple testing (e.g., family-wise error rate (FWER) control). To address this issue, we developed BNPower, a user-friendly power calculation tool for data-driven network analysis. BNPower utilizes simulation analysis, taking into account the complexity of the data-driven network analysis model. We have implemented efficient computational strategies to facilitate data-driven network analysis, including subnetwork extraction and permutation tests for controlling FWER, while maintaining low computational costs. The toolkit, which includes a graphical user interface and source codes, is publicly available at the following GitHub repository: https://github.com/bichuan0419/brain_connectome_power_tool