Hierarchical Bayesian inference for community detection and connectivity of functional brain networks

A team of researchers led by Lingbin Bian and Nizhuan Wang has published a new paper titled 'Hierarchical Bayesian inference for community detection and connectivity of functional brain networks.' The work introduces a novel multilayer community detection method based on a Bayesian latent block model (LBM). This approach is designed to overcome a key limitation in most existing fMRI analysis methods, which fail to properly account for variability between individual subjects. The new model can robustly detect the community structure of weighted functional brain networks without requiring a pre-specified number of communities, and it retains individual-level network variability, providing a more nuanced view of brain organization.

For validation, the team created a new community structure-based multivariate Gaussian generative model to simulate synthetic brain signals. Their simulation study confirmed that the community memberships estimated by their hierarchical Bayesian inference were consistent with the predefined node labels. The method was then rigorously tested using real-world data: split-half reproducibility analysis on working memory task fMRI data from 100 unrelated healthy subjects from the Human Connectome Project. The results, using both synthetic and real data, demonstrate that the proposed hierarchical Bayesian method is more accurate and reliable compared to commonly used (multilayer) modularity models, marking a significant step forward in the statistical toolkit for neuroscience.