Aspectes metodològics i aplicacions de la modelització del temps de supervivència multivariant mitjançant models mixtes

Els estudis de supervivència s'interessen pel temps que passa des de l'inici de l'estudi (diagnòstic de la malaltia, inici del tractament,...) fins que es produeix l'esdeveniment d'interès (mort, curació, millora,...). No obstant això, moltes vegades aquest esdeveniment s'observa més d'una vegada en un mateix individu durant el període de seguiment (dades de supervivència multivariant). En aquest cas, és necessari utilitzar una metodologia diferent a la utilitzada en l'anàlisi de supervivència estàndard. El principal problema que l'estudi d'aquest tipus de dades comporta és que les observacions poden no ser independents. Fins ara, aquest problema s'ha solucionat de dues maneres diferents en funció de la variable dependent. Si aquesta variable segueix una distribució de la família exponencial s'utilitzen els models lineals generalitzats mixtes (GLMM); i si aquesta variable és el temps, variable amb una distribució de probabilitat no pertanyent a aquesta família, s'utilitza l'anàlisi de supervivència multivariant. El que es pretén en aquesta tesis és unificar aquests dos enfocs, és a dir, utilitzar una variable dependent que sigui el temps amb agrupacions d'individus o d'observacions, a partir d'un GLMM, amb la finalitat d'introduir nous mètodes pel tractament d'aquest tipus de dades.

The conditional performance of euro bond funds: evidence from Portugal during the debt crisis

This paper evaluates the performance of a survivorship bias-free data set of Portuguese funds investing in Euro-denominated bonds by using conditional models that consider the public information available to investors when the returns are generated. We find that bond funds underperform the market significantly and by an economically relevant magnitude. This underperformance cannot be explained by the expenses they charge. Our findings support the use of conditional performance evaluation models, since we find strong evidence of both time-varying risk and performance, dependent on the slope of the term structure and the inverse relative wealth variables. We also show that survivorship bias has a significant impact on performance estimates. Furthermore, during the European debt crisis, bond fund managers performed significantly better than in non-crisis periods and were able to achieve neutral performance. This improved performance throughout the crisis seems to be related to changes in funds’ investment styles.

Investor sentiment as conditioning information in asset pricing

This paper assesses whether incorporating investor sentiment as conditioning information in asset-pricing models helps capture the impacts of the size, value, liquidity and momentum effects on risk-adjusted returns of individual stocks. We use survey sentiment measures and a composite index as proxies for investor sentiment. In our conditional framework, the size effect becomes less important in the conditional CAPM and is no longer significant in all the other models examined. Furthermore, the conditional models often capture the value, liquidity and momentum effects.

Identification of renin progenitors in the mouse bone marrow that give rise to B-cell leukaemia

cell of origin and triggering events for leukaemia are mostly unknown. Here we show that the bone marrow contains a progenitor that expresses renin throughout development and possesses a B-lymphocyte pedigree. This cell requires RBP-J to differentiate. Deletion of RBP-J in these renin-expressing progenitors enriches the precursor B-cell gene programme and constrains lymphocyte differentiation, facilitated by H3K4me3 activating marks in genes that control the pre-B stage. Mutant cells undergo neoplastic transformation, and mice develop a highly penetrant B-cell leukaemia with multi-organ infiltration and early death. These reninexpressing cells appear uniquely vulnerable as other conditional models of RBP-J deletion do not result in leukaemia. The discovery of these unique renin progenitors in the bone marrow and the model of leukaemia described herein may enhance our understanding of normal and neoplastic haematopoiesis.

Maximum Entropy Discrimination

We present a general framework for discriminative estimation based on the maximum entropy principle and its extensions. All calculations involve distributions over structures and/or parameters rather than specific settings and reduce to relative entropy projections. This holds even when the data is not separable within the chosen parametric class, in the context of anomaly detection rather than classification, or when the labels in the training set are uncertain or incomplete. Support vector machines are naturally subsumed under this class and we provide several extensions. We are also able to estimate exactly and efficiently discriminative distributions over tree structures of class-conditional models within this framework. Preliminary experimental results are indicative of the potential in these techniques.

Modelling structural zeros in compositional data

This analysis was stimulated by the real data analysis problem of household expenditure data. The full dataset contains expenditure data for a sample of 1224 households. The expenditure is broken down at 2 hierarchical levels: 9 major levels (e.g. housing, food, utilities etc.) and 92 minor levels. There are also 5 factors and 5 covariates at the household level. Not surprisingly, there are a small number of zeros at the major level, but many zeros at the minor level. The question is how best to model the zeros. Clearly, models that try to add a small amount to the zero terms are not appropriate in general as at least some of the zeros are clearly structural, e.g. alcohol/tobacco for households that are teetotal. The key question then is how to build suitable conditional models. For example, is the sub-composition of spending excluding alcohol/tobacco similar for teetotal and non-teetotal households? In other words, we are looking for sub-compositional independence. Also, what determines whether a household is teetotal? Can we assume that it is independent of the composition? In general, whether teetotal will clearly depend on the household level variables, so we need to be able to model this dependence. The other tricky question is that with zeros on more than one component, we need to be able to model dependence and independence of zeros on the different components. Lastly, while some zeros are structural, others may not be, for example, for expenditure on durables, it may be chance as to whether a particular household spends money on durables within the sample period. This would clearly be distinguishable if we had longitudinal data, but may still be distinguishable by looking at the distribution, on the assumption that random zeros will usually be for situations where any non-zero expenditure is not small. While this analysis is based on around economic data, the ideas carry over to many other situations, including geological data, where minerals may be missing for structural reasons (similar to alcohol), or missing because they occur only in random regions which may be missed in a sample (similar to the durables)

Discrete and continuous compositions

This paper examines a dataset which is modeled well by the Poisson-Log Normal process and by this process mixed with Log Normal data, which are both turned into compositions. This generates compositional data that has zeros without any need for conditional models or assuming that there is missing or censored data that needs adjustment. It also enables us to model dependence on covariates and within the composition

Análise à volatilidade do mercado da arte

Um olhar breve para a história financeira internacional denota que, em períodos de crise ou até mesmo de depressão das principais economias mundiais, os investidores tendem a proteger os seus investimentos, muito para além dos usuais activos financeiros, como o caso das acções, obrigações, entre outros. A principal razão para tal é que, em termos empíricos, verifica-se que estes activos referidos apresentam uma elevada volatilidade, especialmente em momentos de grande turbulência nos mercados financeiros, em virtude da incerteza quanto ao futuro das economias mundiais. Assim sendo, a presente investigação realiza uma análise em torno da previsão de volatilidade dos activos associados a arte e, assim, pretende comparar com a volatilidade existente em torno dos índices ou mercados financeiros. Para tal, serão adoptados os modelos de heterocedasticidade condicional, com a finalidade de previsão de volatilidade marginal ou incondicional. A análise efectuada baseou-se na comparação da volatilidade marginal dos índices S&P 500 e DJ Euro Stoxx 50, representativos dos activos financeiros, face à volatilidade marginal das principais empresas (Christie´s e Sotheby´s) e do principal índice de arte (ArtPrice Global Index), representativos dos activos associados à arte. Os resultados evidenciam uma diferença significativa entre as volatilidades marginais ou incondicionais previstas, resultando numa menor volatilidade prevista incondicional dos activos de arte face aos activos financeiros.

Modelagem em análise de sobrevivencia com eventos recorrentes aplicada a dados da área médica

Pós-graduação em Biometria - IBB

Konditionale Modellsysteme zur Untersuchung der ERBB2-induzierten Tumorgenese

Konditionale Modellsysteme zur Untersuchung der ERBB2-induzierten Tumorgenese Die Rezeptor-Tyrosinkinase ERBB2 ist in einer Vielzahl epithelialer Tumore, wie Mamma- und Ovarialkarzinomen, überexprimiert. Diese erhöhte Expression korreliert mit aggressivem Tumorwachstum, verstärkter Metastasierung und schlechter Prognose für den Patienten. Zur genaueren Untersuchung molekularer Mechanismen, die zur Tumorentstehung infolge der ERBB2-Überexpression führen, wurden im Rahmen dieser Arbeit mit Hilfe des Tet-Systems induzierbare MCF-7 Zelllinien generiert. Diese exprimieren bei Gabe von Doxyzyklin ERBB2 bzw. die zum humanen ERBB2 homologe und durch Punktmutation onkogen aktivierte Rattenvariante NeuT. Nachdem die stringente Regulierbarkeit durch Doxyzyklin für die untersuchten Zellklone gezeigt werden konnte, stellte sich bei der Charakterisierung der Zelllinien heraus, dass die Induktion von ERBB2 erstaunlicherweise nicht zur Proliferation der Zellen, sondern zum Wachstumsarrest führt. Bei der Untersuchung verschiedener Zellzyklusregulatoren konnte dieser Zellzyklusarrest dem CDK-Inhibitor P21 zugeordnet werden, dessen Expression durch ERBB2 induziert wird. In P21-Antisense-Experimenten konnte nachgewiesen werden, dass P21 eine Schlüsselrolle beim ERBB2-induzierten Zellzyklusarrest spielt. Neben der Induktion von P21 und der daraus resultierenden Wachstumsinhibition zeigten die Zellen starke morphologische Veränderungen und waren positiv beim Nachweis der Seneszenz-assoziierten -Galaktosidase. Erstmals konnte gezeigt werden, dass die Induktion des Onkogens ERBB2 nicht zur Proliferation, sondern zur Aktivierung eines verfrühten Seneszenz-Programms führt, welches der Zelle Schutz gegen die Onkogeneinwirkung bietet. Bei der Untersuchung verschiedener Signaltransduktionskaskaden mit Inhibitormolekülen konnte die Aktivierung dieses Seneszenz-Programms der Stress-aktivierten Proteinkinase P38 zugeordnet werden. Zur Identifizierung von Genen, die für die ERBB2-induzierte Tumorgenese relevant sind, wurde die differenzielle Genexpression eines NeuT-Klons nach 8- bzw. 48-stündiger Induktion mit Doxyzyklin in einem cDNA-Array untersucht. Dabei zeigte sich eine besonders starke Induktion von Integrin 5 und Integrin 1, die zusammen den Fibronektinrezeptor bilden. Der funktionale Nachweis des Rezeptors in einem Adhäsionsassay demonstrierte ein stark erhöhtes Adhäsionsverhalten ERBB2-überexprimierender Zellen an Fibronektin. Bei der Untersuchung von Mamma-, Ovarial- und Endometriumkarzinomen konnte die Expression von ERBB2 mit der von Integrin 5 korreliert werden. Diese Ergebnisse machen Integrin 5 zu einem potenziellen neuen Tumormarker und Therapieziel in ERBB2-überexprimierenden Tumoren. Ein weiteres interessantes Gen, das sich im Array durch ERBB2 überexprimiert zeigte, war die Matrix-Metalloproteinase MMP-9. In einem Zymografieassay konnte die erhöhte Gelatinaseaktivität von MMP-9 in Dox-induzierten Zellen nachgewiesen werden. Der Einsatz verschiedener Signaltransduktionsinhibitoren ergab, dass auch die ERBB2-induzierte Expression von MMP-9 über die Aktivierung von P38 läuft. Bei der Suche nach weiteren MMPs, die für die ERBB2-induzierte Tumorgenese relevant sein könnten, wurde MMP-13 untersucht. Erstmals konnte gezeigt werden, dass diese Matrix-Metalloproteinase von ERBB2 induziert wird. Dieser interessante Befund wurde auch in einem anderen Zellmodell in NIH3T3 Mausfibroblasten verifiziert. Durch ihre Matrix-degradierenden Eigenschaften sind MMPs potente „Werkzeuge“ für Tumorzellen und stellen ein wichtiges Ziel zur Unterbindung der Invasion und Metastasierung dieser Zellen dar. Neben den Zellkulturarbeiten wurden im Rahmen dieser Dissertation transgene Responder-Mäuse generiert, die NeuT unter Kontrolle eines Tet-responsiven Promotors exprimieren. Von vier transgenen Gründerlinien zeigten zwei eine unerwünschte, basale NeuT-Expression, für die beiden anderen Linien konnte sowohl in MEF-Assays, als auch nach Kreuzung mit rtTA- bzw. tTA-Effektor-Mäusen eine Dox-abhängige Regulation des Transgens gezeigt werden. Die Tiere dieser Linien sollen in Zukunft mit Effektor-Mäusen gepaart werden, die den rtTA bzw. tTA spezifisch in für die ERBB2-Tumorgenese relevanten Geweben, wie Ovarial- oder Lungenepithelzellen, exprimieren. So können individuelle Tumormodelle für die verschiedenen epithelialen Tumore, bei denen die Überexpression von ERBB2 von Bedeutung ist, entwickelt und untersucht werden.

Modelação e previsão de um índice financeiro (FTSE 100)

A presente dissertação visa uma aplicação de séries temporais, na modelação do índice financeiro FTSE100. Com base na série de retornos, foram estudadas a estacionaridade através do teste Phillips-Perron, a normalidade pelo Teste Jarque-Bera, a independência analisada pela função de autocorrelação e pelo teste de Ljung-Box, e utilizados modelos GARCH, com a finalidade de modelar e prever a variância condicional (volatilidade) da série financeira em estudo. As séries temporais financeiras apresentam características peculiares, revelando períodos mais voláteis do que outros. Esses períodos encontram-se distribuídos em clusters, sugerindo um grau de dependência no tempo. Atendendo à presença de tais grupos de volatilidade (não linearidade), torna-se necessário o recurso a modelos heterocedásticos condicionais, isto é, modelos que consideram que a variância condicional de uma série temporal não é constante e dependente do tempo. Face à grande variabilidade das séries temporais financeiras ao longo do tempo, os modelos ARCH (Engle, 1982) e a sua generalização GARCH (Bollerslev, 1986) revelam-se os mais adequados para o estudo da volatilidade. Em particular, estes modelos não lineares apresentam uma variância condicional aleatória, sendo possível, através do seu estudo, estimar e prever a volatilidade futura da série. Por fim, é apresentado o estudo empírico que se baseia numa proposta de modelação e previsão de um conjunto de dados reais do índice financeiro FTSE100.

An infinite-horizon robust filter for uncertain hidden Markov Models with conditional relative entropy constraints

We consider a robust filtering problem for uncertain discrete-time, homogeneous, first-order, finite-state hidden Markov models (HMMs). The class of uncertain HMMs considered is described by a conditional relative entropy constraint on measures perturbed from a nominal regular conditional probability distribution given the previous posterior state distribution and the latest measurement. Under this class of perturbations, a robust infinite horizon filtering problem is first formulated as a constrained optimization problem before being transformed via variational results into an unconstrained optimization problem; the latter can be elegantly solved using a risk-sensitive information-state based filtering.

Novel conditional mouse models for targeted kidney research : Emphasis on the analysis of the biological function of nephrin

Nephrin is a transmembrane protein belonging to the immunoglobulin superfamily and is expressed primarily in the podocytes, which are highly differentiated epithelial cells needed for primary urine formation in the kidney. Mutations leading to nephrin loss abrogate podocyte morphology, and result in massive protein loss into urine and consequent early death in humans carrying specific mutations in this gene. The disease phenotype is closely replicated in respective mouse models. The purpose of this thesis was to generate novel inducible mouse-lines, which allow targeted gene deletion in a time and tissue-specific manner. A proof of principle model for succesful gene therapy for this disease was generated, which allowed podocyte specific transgene replacement to rescue gene deficient mice from perinatal lethality. Furthermore, the phenotypic consequences of nephrin restoration in the kidney and nephrin deficiency in the testis, brain and pancreas in rescued mice were investigated. A novel podocyte-specific construct was achieved by using standard cloning techniques to provide an inducible tool for in vitro and in vivo gene targeting. Using modified constructs and microinjection procedures two novel transgenic mouse-lines were generated. First, a mouse-line with doxycycline inducible expression of Cre recombinase that allows podocyte-specific gene deletion was generated. Second, a mouse-line with doxycycline inducible expression of rat nephrin, which allows podocyte-specific nephrin over-expression was made. Furthermore, it was possible to rescue nephrin deficient mice from perinatal lethality by cross-breeding them with a mouse-line with inducible rat nephrin expression that restored the missing endogenous nephrin only in the kidney after doxycycline treatment. The rescued mice were smaller, infertile, showed genital malformations and developed distinct histological abnormalities in the kidney with an altered molecular composition of the podocytes. Histological changes were also found in the testis, cerebellum and pancreas. The expression of another molecule with limited tissue expression, densin, was localized to the plasma membranes of Sertoli cells in the testis by immunofluorescence staining. Densin may be an essential adherens junction protein between Sertoli cells and developing germ cells and these junctions share similar protein assembly with kidney podocytes. This single, binary conditional construct serves as a cost- and time-efficient tool to increase the understanding of podocyte-specific key proteins in health and disease. The results verified a tightly controlled inducible podocyte-specific transgene expression in vitro and in vivo as expected. These novel mouse-lines with doxycycline inducible Cre recombinase and with rat nephrin expression will be useful for conditional gene targeting of essential podocyte proteins and to study in detail their functions in the adult mice. This is important for future diagnostic and pharmacologic development platforms.

Assessment of conditional moment closure models of turbulent autoignition using DNS data

A systematic assessment of the submodels of conditional moment closure (CMC) formalism for the autoignition problem is carried out using direct numerical simulation (DNS) data. An initially non-premixed, n-heptane/air system, subjected to a three-dimensional, homogeneous, isotropic, and decaying turbulence, is considered. Two kinetic schemes, (1) a one-step and (2) a reduced four-step reaction mechanism, are considered for chemistry An alternative formulation is developed for closure of the mean chemical source term , based on the condition that the instantaneous fluctuation of excess temperature is small. With this model, it is shown that the CMC equations describe the autoignition process all the way up to near the equilibrium limit. The effect of second-order terms (namely, conditional variance of temperature excess sigma(2) and conditional correlations of species q(ij)) in modeling is examined. Comparison with DNS data shows that sigma(2) has little effect on the predicted conditional mean temperature evolution, if the average conditional scalar dissipation rate is properly modeled. Using DNS data, a correction factor is introduced in the modeling of nonlinear terms to include the effect of species fluctuations. Computations including such a correction factor show that the species conditional correlations q(ij) have little effect on model predictions with a one-step reaction, but those q(ij) involving intermediate species are found to be crucial when four-step reduced kinetics is considered. The "most reactive mixture fraction" is found to vary with time when a four-step kinetics is considered. First-order CMC results are found to be qualitatively wrong if the conditional mean scalar dissipation rate is not modeled properly. The autoignition delay time predicted by the CMC model compares excellently with DNS results and shows a trend similar to experimental data over a range of initial temperatures.