UCSC-SOE-20-11: Bayesian Supervised Clustering of Undirected Networks for Understanding Impacts of the Brain Network on Human Creativity

Sharmistha Guha and Rajarshi Guhaniyogi
10/26/2020 09:12 PM
This article develops a flexible relationship between a measure of creative
achievement, the Creative Achievement Questionnaire (CAQ), and the
brain network of subjects from a brain connectome dataset obtained using
a diffusion weighted magnetic resonance imaging (DWI) technique. Undirected
brain networks are often visualized using symmetric adjacency matrices,
with row and column indices of the matrix representing regions of
interest (ROI), and a cell entry signifying the estimated number of fiber bundles
connecting the corresponding row and column ROIs. Motivated by earlier
studies on the differences in the relationship between brain connectivity
networks and phenotypic traits for different groups of individuals, this article
aims to cluster individuals according to the shared relationships of their
brain networks and creativity. Additionally, scientific interest lies in identifying
ROIs in the human brain significantly associated with creative achievement
in each cluster of subjects. To address these questions, we propose a
novel Bayesian mixture modeling framework with an undirected network response
and scalar predictors. The symmetric matrix coefficients corresponding
to the scalar predictors of interest in each mixture component are embedded
with low-rankness and group sparsity within the low-rank structure.
Being a principled Bayesian framework allows us to precisely characterize
the uncertainty in detecting significant network nodes in each cluster. Empirical
results in various simulation scenarios illustrate substantial inferential
gains of the proposed framework in comparison with competitors. Analysis
of the brain connectome data with the proposed model reveals interesting insights
into the brain regions significantly related to creative achievement in
each cluster of individuals.