Les modèles de Markov d’ordre 2 (HMM2) sont des modèles stochastiques qui ont démontré leur efficacité dans l’exploration de séquences génomiques. Cette thèse explore l’intérêt de modèles de différents types (M1M2, M2M2, M2M0) ainsi que leur couplage à des méthodes combinatoires pour segmenter les génomes bactériens sans connaissances a priori du contenu génétique. Ces approches ont été appliquées à deux modèles bactériens afin d’en valider la robustesse : Streptomyces coelicolor et Streptococcus thermophilus. Ces espèces bactériennes présentent des caractéristiques génomiques très distinctes (composition, taille du génome) en lien avec leur écosystème spécifique : le sol pour les S. coelicolor et le milieu lait pour S. thermophilus / Second-order Hidden Markov Models (HMM2) are stochastic processes with a high efficiency in exploring bacterial genome sequences. Different types of HMM2 (M1M2, M2M2, M2M0) combined to combinatorial methods were developed in a new approach to discriminate genomic regions without a priori knowledge on their genetic content. This approach was applied on two bacterial models in order to validate its achievements: Streptomyces coelicolor and Streptococcus thermophilus. These bacterial species exhibit distinct genomic traits (base composition, global genome size) in relation with their ecological niche: soil for S. coelicolor and dairy products for S. thermophilus. In S. coelicolor, a first HMM2 architecture allowed the detection of short discrete DNA heterogeneities (5-16 nucleotides in size), mostly localized in intergenic regions. The application of the method on a biologically known gene set, the SigR regulon (involved in oxidative stress response), proved the efficiency in identifying bacterial promoters. S. coelicolor shows a complex regulatory network (up to 12% of the genes may be involved in gene regulation) with more than 60 sigma factors, involved in initiation of transcription. A classification method coupled to a searching algorithm (i.e. R’MES) was developed to automatically extract the box1-spacer-box2 composite DNA motifs, structure corresponding to the typical bacterial promoter -35/-10 boxes. Among the 814 DNA motifs described for the whole S. coelicolor genome, those of sigma factors (B, WhiG) could be retrieved from the crude data. We could show that this method could be generalized by applying it successfully in a preliminary attempt to the genome of Bacillus subtilis

Esta dissertação de mestrado apresenta os resultados de uma investigação sobre os métodos de co-estimativa comumente utilizados em geoestatística. Estes métodos são: cokrigagem ordinária; cokrigagem colocalizada e krigagem com deriva externa. Além disso, a krigagem ordinária foi considerada apenas a título de ilustração como esse método trabalha quando a variável primária estiver pobremente amostrada. Como sabemos, os métodos de co-estimativa dependem de uma variável secundária amostrada sobre o domínio a ser estimado. Adicionalmente, esta variável deveria apresentar correlação linear com a variável principal ou variável primária. Geralmente, a variável primária é pobremente amostrada enquanto a variável secundária é conhecida sobre todo o domínio a ser estimado. Por exemplo, em exploração petrolífera, a variável primária é a porosidade medida em amostras de rocha retiradas de testemunhos e a variável secundária é a amplitude sísmica derivada de processamento de dados de reflexão sísmica. É importante mencionar que a variável primária e a variável secundária devem apresentar algum grau de correlação. Contudo, nós não sabemos como eles funcionam dependendo do grau de correlação. Esta é a questão. Assim, testamos os métodos de co-estimativa para vários conjuntos de dados apresentando diferentes graus de correlação. Na verdade, esses conjuntos de dados foram gerados em computador baseado em algoritmos de transformação de dados. Cinco valores de correlação foram considerados neste estudo: 0,993, 0,870, 0,752, 0,588 e 0,461. A cokrigagem colocalizada foi o melhor método entre todos testados. Este método tem um filtro interno que é aplicado no cálculo do peso da variável secundária, que por sua vez depende do coeficiente de correlação. De fato, quanto maior o coeficiente de correlação, maior é o peso da variável secundária. Então isso significa que este método funciona mesmo quando o coeficiente de correlação entre a variável primária e a variável secundária é baixo. Este é o resultado mais impressionante desta pesquisa. / This master dissertation presents the results of a survey into co-estimation methods commonly used in geostatistics. These methods are ordinary cokriging, collocated cokriging and kriging with an external drift. Besides that ordinary kriging was considered just to illustrate how it does work when the primary variable is poorly sampled. As we know co-estimation methods depend on a secondary variable sampled over the estimation domain. Moreover, this secondary variable should present linear correlation with the main variable or primary variable. Usually the primary variable is poorly sampled whereas the secondary variable is known over the estimation domain. For instance in oil exploration the primary variable is porosity as measured on rock samples gathered from drill holes and the secondary variable is seismic amplitude derived from processing seismic reflection data. It is important to mention that primary and secondary variables must present some degree of correlation. However, we do not know how they work depending on the correlation coefficient. That is the question. Thus, we have tested co-estimation methods for several data sets presenting different degrees of correlation. Actually, these data sets were generated in computer based on some data transform algorithms. Five correlation values have been considered in this study: 0.993; 0.870; 0.752; 0.588 and 0.461. Collocated simple cokriging was the best method among all tested. This method has an internal filter applied to compute the weight for the secondary variable, which in its turn depends on the correlation coefficient. In fact, the greater the correlation coefficient the greater the weight of secondary variable is. Then it means this method works even when the correlation coefficient between primary and secondary variables is low. This is the most impressive result that came out from this research.

Un individu est dit consanguin si ses parents sont apparentés et s’il existe donc dans sa généalogie au moins une boucle de consanguinité aboutissant à un ancêtre commun. Le coefficient de consanguinité de l’individu est par définition la probabilité pour qu’à un point pris au hasard sur le génome, l’individu ait reçu deux allèles identiques par descendance qui proviennent d’un seul allèle présent chez un des ancêtres communs. Ce coefficient de consanguinité est un paramètre central de la génétique qui est utilisé en génétique des populations pour caractériser la structure des populations, mais également pour rechercher des facteurs génétiques impliqués dans les maladies. Le coefficient de consanguinité était classiquement estimé à partir des généalogies, mais des méthodes ont été développées pour s’affranchir des généalogies et l’estimer à partir de l’information apportée par des marqueurs génétiques répartis sur l’ensemble du génome.Grâce aux progrès des techniques de génotypage haut-débit, il est possible aujourd’hui d’obtenir les génotypes d’un individu sur des centaines de milliers de marqueurs et d’utiliser ces méthodes pour reconstruire les régions d’identité par descendance sur son génome et estimer un coefficient de consanguinité génomique. Il n’existe actuellement pas de consensus sur la meilleure stratégie à adopter sur ces cartes denses de marqueurs en particulier pour gérer les dépendances qui existent entre les allèles aux différents marqueurs (déséquilibre de liaison). Dans cette thèse, nous avons évalué les différentes méthodes disponibles à partir de simulations réalisées en utilisant de vraies données avec des schémas de déséquilibre de liaison réalistes. Nous avons montré qu’une approche intéressante consistait à générer plusieurs sous-cartes de marqueurs dans lesquelles le déséquilibre de liaison est minimal, d’estimer un coefficient de consanguinité sur chacune des sous-cartes par une méthode basée sur une chaîne de Markov cachée implémentée dans le logiciel FEstim et de prendre comme estimateur la médiane de ces différentes estimations. L’avantage de cette approche est qu’elle est utilisable sur n’importe quelle taille d’échantillon, voire sur un seul individu, puisqu’elle ne demande pas d’estimer les déséquilibres de liaison. L’estimateur donné par FEstim étant un estimateur du maximum de vraisemblance, il est également possible de tester si le coefficient de consanguinité est significativement différent de zéro et de déterminer la relation de parenté des parents la plus vraisemblable parmi un ensemble de relations. Enfin, en permettant l’identification de régions d’homozygoties communes à plusieurs malades consanguins, notre stratégie peut permettre l’identification des mutations récessives impliquées dans les maladies monogéniques ou multifactorielles.Pour que la méthode que nous proposons soit facilement utilisable, nous avons développé le pipeline, FSuite, permettant d’interpréter facilement les résultats d’études de génétique de populations et de génétique épidémiologique comme illustré sur le panel de référence HapMap III, et sur un jeu de données cas-témoins de la maladie d’Alzheimer. / An individual is said to be inbred if his parents are related and if his genealogy contains at least one inbreeding loop leading to a common ancestor. The inbreeding coefficient of an individual is defined as the probability that the individual has received two alleles identical by descent, coming from a single allele present in a common ancestor, at a random marker on the genome. The inbreeding coefficient is a central parameter in genetics, and is used in population genetics to characterize the population structure, and also in genetic epidemiology to search for genetic factors involved in recessive diseases.The inbreeding coefficient was traditionally estimated from genealogies, but methods have been developed to avoid genealogies and to estimate this coefficient from the information provided by genetic markers distributed along the genome.With the advances in high-throughput genotyping techniques, it is now possible to genotype hundreds of thousands of markers for one individual, and to use these methods to reconstruct the regions of identity by descent on his genome and estimate a genomic inbreeding coefficient. There is currently no consensus on the best strategy to adopt with these dense marker maps, in particular to take into account dependencies between alleles at different markers (linkage disequilibrium).In this thesis, we evaluated the different available methods through simulations using real data with realistic patterns of linkage disequilibrium. We highlighted an interesting approach that consists in generating several submaps to minimize linkage disequilibrium, estimating an inbreeding coefficient of each of the submaps based on a hidden Markov method implemented in FEstim software, and taking as estimator the median of these different estimates. The advantage of this approach is that it can be used on any sample size, even on an individual, since it requires no linkage disequilibrium estimate. FEstim is a maximum likelihood estimator, which allows testing whether the inbreeding coefficient is significantly different from zero and determining the most probable mating type of the parents. Finally, through the identification of homozygous regions shared by several consanguineous patients, our strategy permits the identification of recessive mutations involved in monogenic and multifactorial diseases.To facilitate the use of our method, we developed the pipeline FSuite, to interpret results of population genetics and genetic epidemiology studies, as shown on the HapMap III reference panel, and on a case-control Alzheimer's disease data.

No description available.

Contexte : Les stratégies pharmacologiques pour traiter la schizophrénie reçoivent une attention croissante due au développement de nouvelles pharmacothérapies plus efficaces, mieux tolérées mais plus coûteuses. La schizophrénie est une maladie chronique présentant différents états spécifiques et définis par leur sévérité. Objectifs : Ce programme de recherche vise à: 1) Évaluer les facteurs associés au risque d'être dans un état spécifique de la schizophrénie, afin de construire les fonctions de risque de la modélisation du cours naturel de la schizophrénie; 2) Développer et valider un modèle de Markov avec microsimulations de Monte-Carlo, afin de simuler l'évolution naturelle des patients qui sont nouvellement diagnostiqués pour la schizophrénie, en fonction du profil individuel des facteurs de risque; 3) Estimer le coût direct de la schizophrénie (pour les soins de santé et autres non reliés aux soins de santé) dans la perspective gouvernementale et simuler l’impact clinique et économique du développement d’un traitement dans une cohorte de patients nouvellement diagnostiqués avec la schizophrénie, suivis pendant les cinq premières années post-diagnostic. Méthode : Pour le premier objectif de ce programme de recherche, un total de 14 320 patients nouvellement diagnostiqués avec la schizophrénie ont été identifiés dans les bases de données de la RAMQ et de Med-Echo. Les six états spécifiques de la schizophrénie ont été définis : le premier épisode (FE), l'état de dépendance faible (LDS), l’état de dépendance élevée (HDS), l’état stable (Stable), l’état de bien-être (Well) et l'état de décès (Death). Pour évaluer les facteurs associés au risque de se trouver dans chacun des états spécifiques de la schizophrénie, nous avons construit 4 fonctions de risque en se basant sur l'analyse de risque proportionnel de Cox pour des risques compétitifs. Pour le deuxième objectif, nous avons élaboré et validé un modèle de Markov avec microsimulations de Monte-Carlo intégrant les six états spécifiques de la schizophrénie. Dans le modèle, chaque sujet avait ses propres probabilités de transition entre les états spécifiques de la schizophrénie. Ces probabilités ont été estimées en utilisant la méthode de la fonction d'incidence cumulée. Pour le troisième objectif, nous avons utilisé le modèle de Markov développé précédemment. Ce modèle inclut les coûts directs de soins de santé, estimés en utilisant les bases de données de la Régie de l'assurance maladie du Québec et Med-Echo, et les coûts directs autres que pour les soins de santé, estimés à partir des enquêtes et publications de Statistique Canada. Résultats : Un total de 14 320 personnes nouvellement diagnostiquées avec la schizophrénie ont été identifiées dans la cohorte à l'étude. Le suivi moyen des sujets était de 4,4 (± 2,6) ans. Parmi les facteurs associés à l’évolution de la schizophrénie, on peut énumérer l’âge, le sexe, le traitement pour la schizophrénie et les comorbidités. Après une période de cinq ans, nos résultats montrent que 41% des patients seront considérés guéris, 13% seront dans un état stable et 3,4% seront décédés. Au cours des 5 premières années après le diagnostic de schizophrénie, le coût direct moyen de soins de santé et autres que les soins de santé a été estimé à 36 701 $ canadiens (CAN) (95% CI: 36 264-37 138). Le coût des soins de santé a représenté 56,2% du coût direct, le coût de l'aide sociale 34,6% et le coût associé à l’institutionnalisation dans les établissements de soins de longue durée 9,2%. Si un nouveau traitement était disponible et offrait une augmentation de 20% de l'efficacité thérapeutique, le coût direct des soins de santé et autres que les soins de santé pourrait être réduit jusqu’à 14,2%. Conclusion : Nous avons identifié des facteurs associés à l’évolution de la schizophrénie. Le modèle de Markov que nous avons développé est le premier modèle canadien intégrant des probabilités de transition ajustées pour le profil individuel des facteurs de risque, en utilisant des données réelles. Le modèle montre une bonne validité interne et externe. Nos résultats indiquent qu’un nouveau traitement pourrait éventuellement réduire les hospitalisations et le coût associé aux établissements de soins de longue durée, augmenter les chances des patients de retourner sur le marché du travail et ainsi contribuer à la réduction du coût de l'aide sociale. / Aim: Pharmacological strategies for schizophrenia have received increasing attention due to the development of new therapies more effective, better tolerated but more expensive. Schizophrenia is a chronic illness with various states of illness. Objectives: This research program aimed: 1) to evaluate the factors associated with the risk of being in a specific state of schizophrenia in order to construct the risk functions of the course of schizophrenia modeling; 2) to develop and validate a Markov model with Monte-Carlo micro-simulations in order to simulate the natural course of patients who have been newly diagnosed with schizophrenic based upon the individual risk factors profile; and 3) to estimate the direct healthcare and non-healthcare cost of schizophrenia and to simulate clinical and economic impact of developing a new treatment, in a cohort of patients newly diagnosed with schizophrenia, over the first 5 years following their diagnosis. Methods: For the first objective of this research program, a total of 14,320 newly diagnosed patients with schizophrenia were identified based on data from the RAMQ and Med-Echo databases. Six disorder states of schizophrenia were defined: first episode (FE), low dependency state (LDS), high dependency state (HDS), Stable state (Stable), Well state (Well) and Death state (Death). To evaluate factors associated to the risk of being in each disease state, we constructed 4 risk functions based on the Cox proportional hazard analysis for competing risks. For the second objective, a Markov model with Monte-Carlo microsimulations with the six specific states of schizophrenia was developed and validated. In the model, each subject had his own probabilities of transition between specific states, which were estimated based on the cumulative incidence function. For the third objected, we used the Markov model we previously developed. The model includes direct healthcare costs estimated from the Régie de l’assurance maladie du Québec and Med-Echo databases and direct non-healthcare costs estimated from the surveys and publications of Statistics Canada. Results: A total of 14,320 individuals were identified in the study cohort as newly diagnosed patients with schizophrenia. The mean follow-up of the subjects was of 4.4 (± 2.6) years. The age, the sex, the schizophrenia treatment, and having comorbidities are factors that are associated with the schizophrenia course. After a five-year period, our results show that 41% of patients will be considered as having recovered, 13% will be in stable condition and 3.4% of patients will have died. The mean direct healthcare and non-healthcare cost of schizophrenia over the first 5 years following diagnosis was estimated $36,701 Canadian (CAN) (95% CI: 36,264 to 37,138). The direct healthcare cost accounted for 56.2% of the total cost, welfare assistance for 34.6% and long term care facilities for 9.2%. On the direct healthcare cost, hospitalisation cost accounted for 64.6%, medical cost for 11.4% and drug-related cost for 24%. In the case where a new treatment with 20% increase of effectiveness will be available, the direct healthcare and non-healthcare costs can be reduced up to 14.2%. Conclusion: We have identified factors associated with the schizophrenia’s specific states, The Markov model we have developed is the first Canadian model incorporating transition probabilities adjusted for individual risk factor profiles using real-life data. The model shows a good internal and external validity. Based on the cost estimates, our results indicate that a new treatment could possibly reduce hospitalization and long-term care facility costs while potentially enabling patients to return to active employment that would in turn contribute to the reduction of the welfare assistance cost.

This thesis presents a cross-layer design and optimization for emerging wideband wireless networks supporting multimedia applications, considering the interactions of the wireless channel characteristics, the physical and link layer protocols, and the user-perceived Quality-of-Service (QoS). As wireless channels are error-prone and broadcast in nature, both the error control mechanisms and the Media Access Control (MAC) protocols are critical for resource utilization and QoS provisioning. How to analyze, design and optimize the high-rate wireless networks by considering the characteristics of the propagation channels and wideband communication technologies is an open, challenging issue. In this thesis, we consider two important wideband wireless systems, the Ultra-Wideband (UWB) and the Orthogonal Frequency-Division Multiplexing (OFDM) systems. First, we propose the packet-level channel models based on Finite State Markov Chains (FSMCs) for the two systems, which present the statistical properties of the propagation channels and the transmission systems. Second, by incorporating the proposed packet-level channel models, we develop analytical frameworks for quantifying the performance of the high-rate wireless networks, combining the channel fading, physical- and link-layer error-control mechanisms and MAC protocols. Third, to mitigate the impact of channel fading and impairments, a cross-layer joint error-control mechanism is proposed. In addition, we also investigate the impact of channel fading on the video streaming applications, and propose a simple admission control algorithm to ensure QoS. As considering the physical-layer characteristics is critical for ensuring QoS and efficiency of resource utilization, the packet-level channel models, cross-layer analytical frameworks, networking protocols and simulation methodologies proposed in this dissertation are essential for future proliferation of high-rate wireless networks.

Nous développons une méthode générique semi-automatique multi-objet de segmentation d'image par coupure de graphe visant les usages médicaux de routine, allant des tâches impliquant quelques objets dans des images 2D, à quelques dizaines dans celles 3D quasi corps entier. La formulation souple de la méthode permet son adaptation simple à une application donnée. En particulier, le modèle d'a priori de proximité que nous proposons, défini à partir des contraintes de paires du plus court chemin sur le graphe d'adjacence des objets, peut facilement être adapté pour tenir compte des relations spatiales entre les objets ciblés dans un problème donné. L'algorithme de segmentation peut être adapté aux besoins de l'application en termes de temps d'exécution et de capacité de stockage à l'aide d'une partition de l'image à segmenter par une tesselation de Voronoï efficace et contrôlable, établissant un bon équilibre entre la compacité des régions et le respect des frontières des objets. Des évaluations et comparaisons qualitatives et quantitatives avec le modèle de Potts standard confirment que notre modèle d'a priori apporte des améliorations significatives dans la segmentation d'objets distincts d'intensités similaires, dans le positionnement précis des frontières des objets ainsi que dans la robustesse de segmentation par rapport à la résolution de partition. L'évaluation comparative de la méthode de partition avec ses concurrentes confirme ses avantages en termes de temps d'exécution et de qualité des partitions produites. Par comparaison avec l'approche appliquée directement sur les voxels de l'image, l'étape de partition améliore à la fois le temps d'exécution global et l'empreinte mémoire du processus de segmentation jusqu'à un ordre de grandeur, sans compromettre la qualité de la segmentation en pratique. / We develop a generic Graph Cut-based semiautomatic multiobject image segmentation method principally for use in routine medical applications ranging from tasks involving few objects in 2D images to fairly complex near whole-body 3D image segmentation. The flexible formulation of the method allows its straightforward adaption to a given application.\linebreak In particular, the graph-based vicinity prior model we propose, defined as shortest-path pairwise constraints on the object adjacency graph, can be easily reformulated to account for the spatial relationships between objects in a given problem instance. The segmentation algorithm can be tailored to the runtime requirements of the application and the online storage capacities of the computing platform by an efficient and controllable Voronoi tessellation clustering of the input image which achieves a good balance between cluster compactness and boundary adherence criteria. Qualitative and quantitative comprehensive evaluation and comparison with the standard Potts model confirm that the vicinity prior model brings significant improvements in the correct segmentation of distinct objects of identical intensity, the accurate placement of object boundaries and the robustness of segmentation with respect to clustering resolution. Comparative evaluation of the clustering method with competing ones confirms its benefits in terms of runtime and quality of produced partitions. Importantly, compared to voxel segmentation, the clustering step improves both overall runtime and memory footprint of the segmentation process up to an order of magnitude virtually without compromising the segmentation quality.

The thesis investigates different exchange rate models and their forecasting performance. The work takes previous literature overview and summarize their findings. Despite the significant amount of papers which were done on the topic of exchange rate forecast, basically none of them cannot find an appropriate model which would outperform a forecast of a simple random walk in every horizon or for any currency pair. However, there are some positive findings in specific cases (e.g. for specific pair or for specific time horizon). The study provides up-to-date analysis of four exchange rates (USD/CZK, USD/ILS, USD/GBP and USD/EUR) for the period of time from January 2000 to August 2013 and analyse forecasting performance of seven exchange rate models (uncovered interest rate parity model, purchasing power parity model, monetary model, monetary model with error correction, Taylor rule model, hidden Markov model and ESTAR model). Although, the results are in advantage of Taylor rule model, especially for the exchange rate of USD/CZK, I cannot prove that the forecasting performance is significantly better than the random walk model. Except of the overall analysis, the work suppose instabilities in the time. Stock and Watson (2003) found that the forecast predictability is not stable over time. As a consequence, the econometric model can give us better forecast than random walk process at some period of time, however at other period, the forecasting ability can be worse than random walk. Based on Fluctuation test of Giacomini and Rossi (2010a) every model is analysed how the out-of-sample forecast ability changes over time.

Transformations between different geodetic reference frames are often performed such that first the transformation parameters are determined from control points. If in the first place we do not know which of the numerous transformation models is appropriate then we can set up a multiple hypotheses test. The paper extends the common method of testing transformation parameters for significance, to the case that also constraints for such parameters are tested. This provides more flexibility when setting up such a test. One can formulate a general model with a maximum number of transformation parameters and specialize it by adding constraints to those parameters, which need to be tested. The proper test statistic in a multiple test is shown to be either the extreme normalized or the extreme studentized Lagrange multiplier. They are shown to perform superior to the more intuitive test statistics derived from misclosures. It is shown how model selection by multiple hypotheses testing relates to the use of information criteria like AICc and Mallows’ Cp, which are based on an information theoretic approach. Nevertheless, whenever comparable, the results of an exemplary computation almost coincide.

Many normal and cancerous cell lines exhibit a stable composition of cells in distinct states which can, e.g., be defined on the basis of cell surface markers. There is evidence that such an equilibrium is associated with stochastic transitions between distinct states. Quantifying these transitions has the potential to better understand cell lineage compositions. We introduce CellTrans, an R package to quantify stochastic cell state transitions from cell state proportion data from fluorescence-activated cell sorting and flow cytometry experiments. The R package is based on a mathematical model in which cell state alterations occur due to stochastic transitions between distinct cell states whose rates only depend on the current state of a cell. CellTrans is an automated tool for estimating the underlying transition probabilities from appropriately prepared data. We point out potential analytical challenges in the quantification of these cell transitions and explain how CellTrans handles them. The applicability of CellTrans is demonstrated on publicly available data on the evolution of cell state compositions in cancer cell lines. We show that CellTrans can be used to (1) infer the transition probabilities between different cell states, (2) predict cell line compositions at a certain time, (3) predict equilibrium cell state compositions, and (4) estimate the time needed to reach this equilibrium. We provide an implementation of CellTrans in R, freely available via GitHub (https://github.com/tbuder/CellTrans).