Bayesian nonparametric modeling of populations of brain networks

In neuroscience there is increasing interest in relating the structural connectivity network of white matter fibers in the human brain to cognitive traits and neuropsychiatric disorders. In fact, there is evidence that the structural brain network is an important driver of variability in cognitive traits and disorders. Recent connectomics pipelines can estimate the brain network based on diffusion tensor imaging and structural MRI. This produces a realization from a network-valued random variable for each individual in a study. I will present recent nonparametric Bayes advances for analyzing network-valued data, and for performing inference on the relationship between brain networks and cognitive traits. These methods are provably flexible, reduce dimension adaptively and can be used for formal inferences on group differences adjusting for multiple comparisons automatically. I will discuss key improvements relative to current approaches and illustrate the methods through application to creative reasoning and Alzheimer’s disease data.