O objetivo deste trabalho é estudar a performance de alguns métodos de comparações múltiplas (MCMs) que ajustam o valor-p quando as estatísticas empregadas nos testes são a log-rank e a Cramér-von Mises, ambas não paramétricas e com estrutura de dependência. A vantagem dos MCMs que ajustam o valor-p é que eles controlam as taxas de erro tipo I e tipo II para cada hipótese, afim de atingir um poder estatístico elevado, mantendo a taxa de erro da família dos testes (FWER) menor ou igual ao nível de significância escolhido. Trabalhamos com o procedimento clássico de Bonferroni e com outros métodos vistos como seu melhoramento, com especial atenção a certos procedimentos derivados do método de Simes que permitem realizar inferências sob as hipóteses individuais. Foi verificado teoricamente que a estatística log-rank pertence à classe multivariada totalmente positiva de ordem 2 (\'MTP IND. 2\'), uma vez que o método de Simes garante o controle da FWER quando as estatísticas dependentes assumem esta condição. O controle da FWER empregando a estatística de Cramér-von Mises foi observado apenas por meio de simulações. Os MCMs foram analisados através de estudos computacionais em modelos discretos e contínuos sob censura com foco no problema de comparar um tratamento versus controle / The aim of this work is to study the performance of some Multiple Comparison Methods (MCMs) that adjust the p-value when the log-rank-type and Cramér-von Mises statistics are used, both nonparametric and with dependency structure. The advantage of these methods is that they control the error rates of type I and type II for each hypothesis in order to achieve high statistical power while keeping the Family Wise Error Rate (FWER) lower or equal than a given significance level. The classical Bonferroni procedure is used as well as others seen as its improvement, with special attention to certain procedures derived from Simes\' method for making inferences on individual hypothesis. It is theoretically proved that the weighted Log-Rank statistics belongs to the multivariate totally positive of order 2 (\'MTP IND. 2\') class, which is needed in order to apply Simes\' method, that guarantees control of the FWER of dependent statistics in this case. The control of the FWER when the Cramér-von Mises statistics is used is only veried by means of computational simulations. The MCMs are also analyzed by means of computational experiments with discrete and continuous data under censoring with focus on the problem of comparisons of treatment versus a control

Cette thèse est motivée par l’analyse des données issues de l’imagerie par résonance magnétique fonctionnelle (IRMf). La nécessité de développer des méthodes capables d’extraire la structure sous-jacente des données d’IRMf constitue un challenge mathématique attractif. A cet égard, nous modélisons les réseaux de connectivité cérébrale par un graphe et nous étudions des procédures permettant d’inférer ce graphe.Plus précisément, nous nous intéressons à l’inférence de la structure d’un modèle graphique non orienté par une procédure de test multiple. Nous considérons deux types de structure, à savoir celle induite par la corrélation et celle induite par la corrélation partielle entre les variables aléatoires. Les statistiques de tests basées sur ces deux dernières mesures sont connues pour présenter une forte dépendance et nous les supposerons être asymptotiquement gaussiennes. Dans ce contexte, nous analysons plusieurs procédures de test multiple permettant un contrôle des arêtes incluses à tort dans le graphe inféré.Dans un premier temps, nous questionnons théoriquement le contrôle du False Discovery Rate (FDR) de la procédure de Benjamini et Hochberg dans un cadre gaussien pour des statistiques de test non nécessairement positivement dépendantes. Nous interrogeons par suite le contrôle du FDR et du Family Wise Error Rate (FWER) dans un cadre gaussien asymptotique. Nous présentons plusieurs procédures de test multiple, adaptées aux tests de corrélations (resp. corrélations partielles), qui contrôlent asymptotiquement le FWER. Nous proposons de plus quelques pistes théoriques relatives au contrôle asymptotique du FDR.Dans un second temps, nous illustrons les propriétés des procédures contrôlant asymptotiquement le FWER à travers une étude sur simulation pour des tests basés sur la corrélation. Nous concluons finalement par l’extraction de réseaux de connectivité cérébrale sur données réelles. / This thesis is motivated by the analysis of the functional magnetic resonance imaging (fMRI). The need for methods to build such structures from fMRI data gives rise to exciting new challenges for mathematics. In this regards, the brain connectivity networks are modelized by a graph and we study some procedures that allow us to infer this graph.More precisely, we investigate the problem of the inference of the structure of an undirected graphical model by a multiple testing procedure. The structure induced by both the correlation and the partial correlation are considered. The statistical tests based on the latter are known to be highly dependent and we assume that they have an asymptotic Gaussian distribution. Within this framework, we study some multiple testing procedures that allow a control of false edges included in the inferred graph.First, we theoretically examine the False Discovery Rate (FDR) control of Benjamini and Hochberg’s procedure in Gaussian setting for non necessary positive dependent statistical tests. Then, we explore both the FDR and the Family Wise Error Rate (FWER) control in asymptotic Gaussian setting. We present some multiple testing procedures, well-suited for correlation (resp. partial correlation) tests, which provide an asymptotic control of the FWER. Furthermore, some first theoretical results regarding asymptotic FDR control are established.Second, the properties of the multiple testing procedures that asymptotically control the FWER are illustrated on a simulation study, for statistical tests based on correlation. We finally conclude with the extraction of cerebral connectivity networks on real data set.

O objetivo deste trabalho é estudar a performance de alguns métodos de comparações múltiplas (MCMs) que ajustam o valor-p quando as estatísticas empregadas nos testes são a log-rank e a Cramér-von Mises, ambas não paramétricas e com estrutura de dependência. A vantagem dos MCMs que ajustam o valor-p é que eles controlam as taxas de erro tipo I e tipo II para cada hipótese, afim de atingir um poder estatístico elevado, mantendo a taxa de erro da família dos testes (FWER) menor ou igual ao nível de significância escolhido. Trabalhamos com o procedimento clássico de Bonferroni e com outros métodos vistos como seu melhoramento, com especial atenção a certos procedimentos derivados do método de Simes que permitem realizar inferências sob as hipóteses individuais. Foi verificado teoricamente que a estatística log-rank pertence à classe multivariada totalmente positiva de ordem 2 (\'MTP IND. 2\'), uma vez que o método de Simes garante o controle da FWER quando as estatísticas dependentes assumem esta condição. O controle da FWER empregando a estatística de Cramér-von Mises foi observado apenas por meio de simulações. Os MCMs foram analisados através de estudos computacionais em modelos discretos e contínuos sob censura com foco no problema de comparar um tratamento versus controle / The aim of this work is to study the performance of some Multiple Comparison Methods (MCMs) that adjust the p-value when the log-rank-type and Cramér-von Mises statistics are used, both nonparametric and with dependency structure. The advantage of these methods is that they control the error rates of type I and type II for each hypothesis in order to achieve high statistical power while keeping the Family Wise Error Rate (FWER) lower or equal than a given significance level. The classical Bonferroni procedure is used as well as others seen as its improvement, with special attention to certain procedures derived from Simes\' method for making inferences on individual hypothesis. It is theoretically proved that the weighted Log-Rank statistics belongs to the multivariate totally positive of order 2 (\'MTP IND. 2\') class, which is needed in order to apply Simes\' method, that guarantees control of the FWER of dependent statistics in this case. The control of the FWER when the Cramér-von Mises statistics is used is only veried by means of computational simulations. The MCMs are also analyzed by means of computational experiments with discrete and continuous data under censoring with focus on the problem of comparisons of treatment versus a control