The problem of identifying connections between nodes in a network model is of fundamental importance in the analysis of brain networks because each node represents a specific brain region that can potentially be connected to other brain regions by means of functional relations; the dynamical behavior of each node can be quantified by adopting a correlation measure among time series. In this contest, the whole set of links between nodes in a network can be represented by means of an adjacency matrix with high dimension, that can be obtained by performing a huge number of simultaneous tests on correlations. In this regard, the Thesis has dealt with the problem of multiple testing in a Bayesian perspective, by examining in depth the “Bayesian False Discovery Rate” (FDR), already defined in Efron, and by introducing the “Bayesian Power” (BP). The behavior of the FDR and BP estimators has been analyzed both with asymptotic theory and with Monte Carlo simulations; furthermore, it has been investigated the robustness of the proposed estimators by simulating specific pattern of dependencies among the p-values associated to the multiple comparisons. Such a multiple testing approach, that allows to control both FDR and BP, has been applyied to a dataset provided by the Milan Center for Neuroscience (NeuroMi). Once selected a sample of 70 participants, classified properly into young subjects and elderly subjects, subject by subject network models have been constructed in order to verify two alternative theories about changes in the pattern of functional connectivity as time goes by, namely the de-differentiation hypothesis versus the localization hypothesis. This objective has been achieved by selecting some proper network measures in order to verify the original hypotheses about the pattern of functional connectivity in the elderly’s group and in the group of young subjects, and by constructing some ad-hoc measures.
|Data di pubblicazione:||17-dic-2015|
|Titolo:||Statistical Network Analysis: a Multiple Testing Approach|
|Settore Scientifico Disciplinare:||SECS-S/01 - STATISTICA|
|Corso di dottorato:||STATISTICA - 11R|
|Citazione:||(2015). Statistical Network Analysis: a Multiple Testing Approach. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2015).|
|Parole Chiave (Inglese):||Multiple hypothesis testing, Bayes False discovery rate, Bayes power, p-values, MRI, Brain network, Network measures|
|Appare nelle tipologie:||07 - Tesi di dottorato Bicocca post 2009|