B1679
Title: Two-sample permutation tests for graphical models and random graphs, with applications to brain connectivity
Authors: Cesare Miglioli - University of Geneva (Switzerland) [presenting]
Pasquale Anthony Della Rosa - Vita-Salute San Raffaele University (Italy)
Maria-Pia Victoria-Feser - University of Geneva (Switzerland)
Stephane Guerrier - University of Geneva (Switzerland)
Abstract: A general three-step procedure is proposed which entails the selection of: i) a model specification, ii) a network representation and iii) a network statistic of interest. Then, a novel class of two-sample Monte Carlo permutation tests is introduced for network data, which are identified by the configuration chosen during the three steps. This general testing framework has the relevant feature, under weak assumptions, of being asymptotically valid and at the same time retaining the exact rejection probability $\alpha$, in finite samples, when the underlying distributions of the two samples are identical. To evaluate the novel procedure, a random sample of $31$ pregnant women are collected who underwent resting-state functional magnetic resonance (rsf-MRI). The $31$ participants were characterized by low-risk (LR), $n=19$ subjects, and high-risk (HR), $m=12$ subjects, for preterm birth (PTB), i.e. any birth occurring before the $37$th week of gestation, based upon a multidimensional assessment and characterization of maternal risk profiles. The results of the permutation test, clearly show the presence of a different fetal brain functional connectivity in HR of PTB pregnancies compared to LR pregnancies. Thus, evidence is provided that altered neurodevelopment, with differential fetal brain connectivity, is not a mere consequence of PTB, instead, it can even anticipate birth.