Weighted brain networks in disease: centrality and entropy in HIV and aging

• Published in: Neurobiology of aging 2014-06, Vol.36 (1), p.401-412
• Main Authors: Thomas, Jewell B., Brier, Matthew R., Ortega, Mario, Benzinger, Tammie L., Ances, Beau M.
• Format: Article
• Language: English
• ISSN: 0197-4580; 1558-1497
• DOI: 10.1016/j.neurobiolaging.2014.06.019

Graph theory models can produce simple, biologically informative metrics of the topology of resting-state functional connectivity (FC) networks. However, typical graph theory approaches model FC relationships between regions (nodes) as unweighted edges, complicating their interpretability in studies of disease or aging. We extended existing techniques and constructed fully-connected weighted graphs for groups of age-matched HIV positive (n=67) and HIV negative (n=77) individuals. We compared test-retest reliability of weighted vs. unweighted metrics in an independent study of healthy individuals (n=22) and found weighted measures to be more stable. We quantified two measures of node centrality (closeness centrality and eigenvector centrality) to capture the relative importance of individual nodes. We also quantified one measure of graph entropy (diversity) to measure the variability in connection strength (edge weights) at each node. HIV was primarily associated with differences in measures of centrality, and age was primarily associated with differences in diversity. HIV and age were associated with divergent measures when evaluated at the whole-graph level, within individual functional networks, and at the level of individual nodes. Graph models may allow us to distinguish previously indistinguishable effects related to HIV and age on FC.