Tesis sobre el tema "Disease progression modeling"

Prognostic studies of progression and mortality in different diseases are essential to understand the role of particular prognostic factors and, thus, improve prognosis and ultimately help selecting appropriate interventions. Yet, such studies often face serious limitations of available data and/or of the existing statistical methods. One difficulty concerns separating the effects of putative prognostic factors on different clinical endpoints or “competing events” such as e.g. disease recurrence vs. recurrence-free death, or death due to disease vs. death due to other causes. This issue becomes even more challenging because data sources, such as cancer registries, often record only the date of death but not the cause of death. This can lead to bias in assessing the role of prognostic factors whose impact on the disease-specific mortality is quite different from their impact on all-cause mortality. It is important, therefore, to use methods that can deal accurately and efficiently with both (i) alternative pathways of disease progression, and (ii) unknown causes of death. The aforementioned challenges are addressed by 3 manuscripts. Previous empirical studies have suggested the potential advantages of using multi-state Markov models, over conventional time-to-event methods, to analyze competing risks and multi-state pathways of disease progression. In the 1st paper, I attempted to systematically assess, through a series of simulations, the performance of Markov models in this context and confirmed the accuracy of both point estimates of the regression coefficients and hypothesis tests. On the other hand, Relative Survival regression models have been developed, in the context of single-endpoint time-to-event analyses, to correct the regression coefficients for the unknown causes of death. Yet, no existing statistical model permits simultaneous combination of the advantages of both (i) Markov multi-state modeling, and (ii) Relative Survival. Therefore, in the
Les études pronostiques sur l'évolution et la mortalité de certaines pathologies sont essentielles pour comprendre le rôle de certains facteurs pronostiques et ainsi, améliorer le pronostic et finalement aider dans le choix des interventions thérapeutiques appropriées. Jusqu'à présent, les études de ce type ont été souvent confrontées à d'importantes limitations dans les données et/ou les méthodes statistiques disponibles. Une des difficultés concerne la discrimination, pour un même facteur pronostique, de ses effets sur différents critères cliniques ou événements concurrents, comme la récidive de la maladie vs le décès sans récidive, ou le décès dû à la pathologie vs le décès dû à d'autres causes. Ce problème devient d'autant plus important que les sources de données, comme les registres, enregistrent souvent uniquement la date de décès mais pas la cause. Ceci peut conduire à des biais dans l'évaluation du rôle des facteurs pronostiques dont l'effet sur la mortalité spécifique dû à la pathologie est différent de celui sur la mortalité toutes causes confondues. Il est donc important d'utiliser des méthodes qui puissent prendre en compte correctement à la fois (i) les différentes évolutions possibles de la pathologie et (ii) l'absence de la connaissance de la cause de décès. Les problèmes méthodologiques mentionnés précédemment sont traités dans 3 articles. Les études empiriques précédentes ont suggéré des avantages potentiels à utiliser les modèles multi-états de Markov à la place des modèles de survie conventionnels dans l'analyse des risques compétitifs et des différentes phases possibles d'évolution d'une pathologie. Le premier article tente d'évaluer méthodiquement, à l'aide de simulations, les performances des modèles de Markov dans ce contexte et confirme l'exactitude à la fois de l'estimation des coefficients de la régression et des tests d'hypothèse. D'un autre coté, les mod

La maladie d'Alzheimer (MA) est une pathologie multi-facette qui peut être surveillée grâce à une grande variété de modalités de données. Cette thèse vise à exploiter des données longitudinales multimodales, principalement des données d'imagerie et des tests cognitifs, pour fournir une description statistique de la progression de la MA et permettre une prévision individuelle de la dégradation future. Les modèles de progression à effet-mixtes de la maladie (DPMs) sont couramment utilisés pour ces tâches. Dans ce contexte, notre première contribution remet en question l'hypothèse fréquente selon laquelle les biomarqueurs suivent des fonctions linéaires ou logistiques au fil du temps, et nous proposons un cadre géométrique qui suppose que les données se trouvent sur une variété et suivent des géodésiques au fil du temps. Nous apprenons la métrique riemannienne de l'espace d'observation et sommes capables de modéliser une plus grande variété de biomarqueurs, sans hypothèses préalables sur la forme de la trajectoire au fil du temps. En utilisant des auto-encodeurs variationnels, nous étendons ensuite ce cadre aux données de neuroimagerie (IRM ou TEP), afin de fournir des modèles de progression en grande dimension qui décrivent les motifs d'altérations structurelles et fonctionnelles du cerveau au cours de la MA. Nous appliquons ensuite cette famille de DPMs à des données réelles afin d'étudier l'hétérogénéité de la progression de la MA, en décrivant l'influence du génotype APOE-e4 et du sexe sur les motifs d'altérations cérébrales. Enfin, nous utilisons ces DPMs avec un ensemble de biomarqueurs d'imagerie et extrait du fluide cérébrospinal pour identifier les combinaisons spécifiques de paramètres qui permettent de prévoir les déclins cognitifs chez les patients à différents stades de la maladie. La thèse démontre que les DPMs peuvent modéliser efficacement la progression de la MA en utilisant une grande variété de données longitudinales multimodales et fournir des informations précieuses sur les manifestations cliniques et la progression de la maladie. Ces résultats peuvent informer la conception d'essais cliniques et faciliter des stratégies de traitement individualisées et plus précises pour les patients atteints de la MA
Alzheimer's disease (AD) is a multi-facet pathology, that can be monitored through a variety of data types. This thesis aims to leverage multimodal longitudinal data, especially imaging scans and cognitive tests, to provide a statistical description of the progression of AD and to enable individual forecasting of future decline. Mixed-effect disease progression models (DPMs) are commonly used for these tasks. In this context, our first contribution questions the frequent assumption that biomarkers follow linear or logistic functions over time, and we propose a geometric framework that assumes the data lie on a manifold and follow geodesics over time. We learn the Riemannian metric of the observation space and are able to model a wider variety of biomarkers, without priors on the shape of the trajectory over time. Using variational auto-encoders, we then extend this framework to neuroimaging data (MRI or PET scans), in order to provide high-dimensional progression models that describe the patterns of structural and functional alterations of the brain over the course of AD. We then apply this family of DPMs to clinical studies data in order to investigate the heterogeneity of AD progression, due to APOE-e4 genotype and sex on patterns of brain alterations. Lastly, we use said DPMs with a set of imaging and fluid biomarkers to identify the specific combinations of input features that best forecast cognitive declines in patients at different stages of the disease. The thesis demonstrates that DPMs can effectively model the progression of AD using a great variety of multimodal longitudinal data and provide valuable insights into the disease's clinical manifestations and progression. These findings can inform clinical trial design and facilitate more accurate prognosis and individualized treatment strategies for patients with AD

Thesis (Ph.D.)--Boston University
Age-related macular degeneration (AMD) is the leading cause of irreversible blindness in people over the age of 50. As many as 50 million people are affected by AMD worldwide and prevalence is expected to continue to rise due to an aging population. There are two forms of the disease, dry (geographic atrophy) and wet (choroidal neovascularization), both of which result in retinal degeneration and central vision loss. Although anti-vascular endothelial growth factor therapies are moderately successful at treating the wet form, there are no treatments currently available for the more common dry form. Pharmacological therapies have been extensively explored for the treatment of dry AMD, but have achieved little success because the pathogenesis underlying AMD is unknown and likely varies among patients . Recently, tissue engineering has emerged as a promising approach to restore function by replacing diseased retinal tissue with healthy retinal pigment epithelium (RPE). While AMD-associated vision loss occurs when photoreceptors degenerate, this process arises as a consequence of earlier RPE dysfunction. In the healthy retina, the RPE acts as a critical regulator of the microenvironment for both photoreceptors and the nearby vasculature. However in AMD, the RPE no longer performs these essential homeostatic functions leading to photoreceptor apoptosis and vision loss. This dissertation describes the development and in vitro characterization of a tissue engineering scaffold for RPE delivery as potential treatment for dry AMD. First, a novel microfabrication-based method termed "pore casting" was developed to produce thin scaffolds with highly controlled pore size, shape, and spacing. Next, human RPE were cultured on pore-cast poly(c-caprolactone) (PCL) scaffolds and compared to cells on track-etched polyester, the standard RPE culture substrate. RPE on porous PCL demonstrated enhanced maturation and function compared to track-etched polyester including improved pigmentation, barrier formation, gene expression, growth factor secretion, and phagocytic degradation. Lastly, this study established a patient-specific method for predicting AMD progression using retinal oxygen concentration. This approach differs from current diagnosis techniques because it uses physiologically-relevant mechanisms rather than generalized clinical associations which have little, if any, prognostic value.

Alzheimer’s disease is a specific form of dementia characterized by the aggregation of Amyloid-β plaques and tau tangles. New research has found that the formation of these aggregates occurs after dysregulation of respiratory activity and the production of radical oxygen species. Proteomic data shows that these changes are also related to unique gene expression patterns. We investigate the impact of these findings on new therapeutic options via metabolic flux analysis of sirtuin stress response pathways and respiratory supercomplex formation. Our results indicate CRISPR Cas-based gene therapy focused on upregulating stable CIII expression, and protective changes in SIRT1 and AMPK expression are potential avenues for therapeutics. This work also highlights the importance of metabolic enzyme activity in maintaining proper respiratory activity.

Given the recognition that disease-modifying therapies should focus on earlier Parkinson's disease stages, trial enrollment based purely on clinical criteria poses significant challenges. The goal herein was to determine the utility of dopamine transporter neuroimaging as an enrichment biomarker in early motor Parkinson's disease clinical trials. Patient-level longitudinal data of 672 subjects with early-stage Parkinson's disease in the Parkinson's Progression Markers Initiative (PPMI) observational study and the Parkinson Research Examination of CEP-1347 Trial (PRECEPT) clinical trial were utilized in a linear mixed-effects model analysis. The rate of worsening in the motor scores between subjects with or without a scan without evidence of dopamine transporter deficit was different both statistically and clinically. The average difference in the change from baseline of motor scores at 24 months between biomarker statuses was -3.16 (90% confidence interval [CI] = -0.96 to -5.42) points. Dopamine transporter imaging could identify subjects with a steeper worsening of the motor scores, allowing trial enrichment and 24% reduction of sample size.

Cette thèse construit des systèmes de coordonnées pour formes, c'est-à-dire des espaces métriques de dimension finie où les formes sont représentées par des vecteurs. Construire de tels systèmes de coordonnées permet de faciliter l'analyse statistique de collections de formes. Notre motivation finale est de prédire et de sous-typer la maladie d'Alzheimer, en se basant notamment sur des marqueurs ainsi extraits de banques d'images médicales du cerveau. Même si de telles banques sont longitudinales, la variabilité qu’elles renferment reste principalement due à la variabilité inter-individuelle importante et normale du cerveau. La variabilité due à la progression d’altérations pathologiques est d'une amplitude beaucoup plus faible. L'objectif central de cette thèse est de développer un système de coordonnées adapté pour l'analyse statistique de banques de données de formes longitudinales, capable de dissocier ces deux sources de variabilité. Comme montré dans la littérature, le transport parallèle peut être exploité pour obtenir une telle dissociation, par exemple en définissant la notion d’exp-parallélisme sur une variété. Utiliser cet outil sur un espace de formes s'accompagne cependant de défis théoriques et calculatoires, relevés dans la première partie de cette thèse. Enfin, si en anatomie computationnelle les espaces de formes sont communément équipés d'une structure de variété, les classes de difféomorphismes sous-jacentes sont le plus souvent construites sans tenir compte des données étudiées. Le dernier objectif majeur de cette thèse est de construire des systèmes de coordonnées de déformations où le paramétrage de ces déformations est adapté aux données d'intérêt
This thesis aims to build coordinate systems for shapes i.e. finite-dimensional metric spaces where shapes are represented by vectors. The goal of building such coordinate systems is to allow and facilitate the statistical analysis of shape data sets. The end-game motivation of our work is to predict and sub-type Alzheimer’s disease, based in part on knowledge extracted from banks of brain medical images. Even if these data banks are longitudinal, their variability remains mostly due to the large and normal inter-individual variability of the brain. The variability due to the progression of pathological alterations is of much smaller amplitude. The central objective of this thesis is to develop a coordinate system adapted for the statistical analysis of longitudinal shape data sets, able to disentangle these two sources of variability. As shown in the literature, the parallel transport operator can be leveraged to achieve this desired disentanglement, for instance by defining the notion of exp-parallel curves on a manifold. Using this tool on shape spaces comes however with theoretical and computational challenges, tackled in the first part of this thesis. Finally, if shape spaces are commonly equipped with a manifold-like structure in the field of computational anatomy, the underlying classes of diffeomorphisms are however most often largely built and parameterized without taking into account the data at hand. The last major objective of this thesis is to build deformation-based coordinate systems where the parameterization of deformations is adapted to the data set of interest

First identified over a decade ago, Astrocyte Elevated Gene-1 (AEG-1) has been studied extensively due to early reports of its overexpression in various cancer cell lines. Research groups all over the globe including our own have since identified AEG-1 overexpression in cancers of diverse lineages including cancers of the liver, colon, skin, prostate, breast, lung, esophagus, neurons and neuronal glia as compared to matched normal tissue. A comprehensive and convincing body of data currently points to AEG-1 as an essential component, critical to the progression and perhaps onset of cancer. AEG-1 is a potent activator of multiple pro-tumorigenic signal transduction pathways such as mitogen-activated protein extracellular kinase (MEK)/ extracellular signal-regulated kinase (ERK), phosphotidyl-inositol-3-kinase (PI3K)/Akt/mTOR, NF-κB and Wnt/β-catenin pathway. In addition, studies show that AEG-1 not only alters global gene and protein expression profiles, it also modulates fundamental intracellular processes, such as transcription, translation and RNA interference in cancer cells most likely by functioning as a scaffold protein. The mechanisms by which AEG-1 is overexpressed in cancer have been studied extensively and it is clear that multiple layers of regulation including genomic amplification, transcriptional, posttranscriptional, and posttranslational controls are involved however; the mechanism by which AEG 1 itself induces its oncogenic effects is still poorly understood. Just as questions remain about the exact role of AEG-1 in carcinogenesis, very little is known about the role of AEG-1 in regulating normal physiological functions in the liver. With the help of the Massey Cancer Center Transgenic/Knockout Mouse Core, our lab has successfully created a germline-AEG-1 knockout mouse (AEG-1-/-) as a model to interrogate AEG-1 function in vivo. Here I present the insights gained from efforts to analyze this novel AEG-1-/- mouse model. Aspects of the physiological functions of AEG-1 will be covered in chapter two wherein details of the characterization of the AEG-1-/- mouse are described including the role of AEG-1 in lipid metabolism. Chapter three discusses novel discoveries about the specific role of AEG-1 in mediating hepatocarcinogenesis by modulating NF-κB, a critical inflammatory pathway. First identified over a decade ago, Astrocyte Elevated Gene-1 (AEG-1) has been studied extensively due to early reports of its overexpression in various cancer cell lines. Research groups all over the globe including our own have since identified AEG-1 overexpression in cancers of diverse lineages including cancers of the liver, colon, skin, prostate, breast, lung, esophagus, neurons and neuronal glia as compared to matched normal tissue. A comprehensive and convincing body of data currently points to AEG-1 as an essential component, critical to the progression and perhaps onset of cancer. AEG-1 is a potent activator of multiple pro-tumorigenic signal transduction pathways such as mitogen-activated protein extracellular kinase (MEK)/ extracellular signal-regulated kinase (ERK), phosphotidyl-inositol-3-kinase (PI3K)/Akt/mTOR, NF-κB and Wnt/β-catenin pathway. In addition, studies show that AEG-1 not only alters global gene and protein expression profiles, it also modulates fundamental intracellular processes, such as transcription, translation and RNA interference in cancer cells most likely by functioning as a scaffold protein. The mechanisms by which AEG-1 is overexpressed in cancer have been studied extensively and it is clear that multiple layers of regulation including genomic amplification, transcriptional, posttranscriptional, and posttranslational controls are involved however; the mechanism by which AEG 1 itself induces its oncogenic effects is still poorly understood. Just as questions remain about the exact role of AEG-1 in carcinogenesis, very little is known about the role of AEG-1 in regulating normal physiological functions in the liver. With the help of the Massey Cancer Center Transgenic/Knockout Mouse Core, our lab has successfully created a germline-AEG-1 knockout mouse (AEG-1-/-) as a model to interrogate AEG-1 function in vivo. Here I present the insights gained from efforts to analyze this novel AEG-1-/- mouse model. Aspects of the physiological functions of AEG-1 will be covered in chapter two wherein details of the characterization of the AEG-1-/- mouse are described including the role of AEG-1 in lipid metabolism. Chapter three discusses novel discoveries about the specific role of AEG-1 in mediating hepatocarcinogenesis by modulating NF-κB, a critical inflammatory pathway.

Diabetes is a major risk factor for the development of cataract [3,14,20,22]. In this thesis, we create a model that allows us to understand the incidence of one disease in the context of another; in particular, cataract in the presence of diabetes. The World Health Organization's Vision 2020 blindness-prevention initiative administers surgeries to remove cataracts, the leading cause of blindness worldwide [24]. One of the geographic areas most impacted by cataract-related blindness is Sub-Saharan Africa. In order to plan the number of surgeries to administer, the World Health Organization uses data on cataract prevalence. However, an estimation of the incidence of cataract is more useful than prevalence data for the purpose of resource planning. In 2012, Dray and Williams developed a method for estimating incidence based on prevalence data [5]. Incidence estimates can be further refined by considering associated risk factors such as diabetes. We therefore extend the Dray and Williams model to include diabetes prevalence when calculating cataract incidence estimates. We explore two possible approaches to our model construction, one a detailed extension, and the other, a simplification of that extension. We provide a discussion comparing the two approaches.

Neurodegenerative disease is an umbrella term for pathologies that primarily damage neurons. As their incidence increases with age it is becoming of a greater concern for the west, due to its aging population. Due to their chronic nature and the difficulty to create reliable and reproducible animal models of these diseases their pathophysiologies are still poorly understood. For all these reasons, a mathematical modelling approach is suggested. The methodology of the work here consisted of identifying the state of the art models that describe the healthy behaviour of cells (e.g. metabolism and ionic regulation) and adapting them for pathological environments. With these models hypotheses provided by clinicians and pathologists were tested. The work focuses on developing models of mechanisms common to neurodegenerative diseases, which include: glutamate excitotoxicity, aquaporin water kinetics, inflammatory complement lysis and acute inflammation. Glutamate excitotoxicity was modelled by creating a compartmental model of glutamate exchange between neurons and astrocytes. This model was the first model of glutamate kinetics validated in an ischaemic stroke context. The aquaporin water kinetics and complement lysis models were developed in the context of the autoimmune disease Neuromyelitis Optica. Through this project a hypothesised trigger for the pathology was confirmed. Additionally, the first model of astrocytic cytotoxic oedema due to complement lysis was developed. Finally, a preventative drug for complement lysis was simulated. Acute inflammation was explored in the context of understanding the potential of chemerin as a pro-resolving cytokine. To that effect, a model of acute inflammation was developed where pro-resolving mechanisms were included. This model was the first to attempt model the effects of an intervention in inflammation. The results indicated that there is a maximum inhibitory effect of chemerin on inflammation. Additionally, two preventive avenues for chronic inflammation were found. With this work, the first attempts of capturing relevant mechanisms of neurodegenerative diseases were presented. These models can now be further developed and adapted to other pathological environments.

The advent of recent high throughput sequencing technologies resulted in an unexplored big data of genomics and transcriptomics that might help to answer various research questions in Parkinson’s disease(PD) progression. While the literature has revealed various predictive models that use longitudinal clinical data for disease progression, there is no predictive model based on RNA-Sequence data of PD patients. This study investigates how to predict the PD Progression for a patient’s next medical visit by capturing longitudinal temporal patterns in the RNA-Seq data. Data provided by Parkinson Progression Marker Initiative (PPMI) includes 423 PD patients with a variable number of visits for a period of 4 years. We propose a predictive model based on a Recurrent Neural Network (RNN) with dense connections. The results show that the proposed architecture is able to predict PD progression from high dimensional RNA-seq data with a Root Mean Square Error (RMSE) of 6.0 and rank-order correlation of (r=0.83, p<0.0001) between the predicted and actual disease status of PD. We show empirical evidence that the addition of dense connections and batch normalization into RNN layers boosts its training and generalization capability.

Quantitative pharmacokinetic-pharmacodynamic and disease progression models are the core of the science of pharmacometrics which has been identified as one of the strategies that can make drug development more effective. To adequately develop and utilize these models one needs to carefully consider the nature of the data, choice of appropriate estimation methods, model evaluation strategies, and, most importantly, the intended use of the model. The general aim of this thesis was to investigate how the use of pharmacometric models can improve the design and analysis of clinical trials within drug development. The development of pharmacometric models for clinical assessment scales in stroke and graded severity events, in this thesis, show the benefit of describing data as close to its true nature as possible, as it increases the predictive abilities and allows for mechanistic interpretations of the models. Performance of three estimation methods implemented in the mixed-effects modeling software NONMEM; 1) Laplace, 2) SAEM, and 3) Importance sampling, applied when modeling repeated time-to-event data, was investigated. The two latter methods are to be preferred if less than approximately half of the individuals experience events. In addition, predictive performance of two validation procedures, internal and external validation, was explored, with internal validation being preferred in most cases. Model-based analysis was compared to conventional methods by the use of clinical trial simulations and the power to detect a drug effect was improved with a pharmacometric design and analysis. Throughout this thesis several examples have shown the possibility of significantly reducing sample sizes in clinical trials with a pharmacometric model-based analysis. This approach will reduce time and costs spent in the development of new drug therapies, but foremost reduce the number of healthy volunteers and patients exposed to experimental drugs.

Dementia progression is characterised by a lengthy pre-symptomatic phase, where pathology may be accumulating, followed by a more rapid decline evidenced by onset of clinical symptoms and eventually functional impairment. The rate of decline in symptoms from cognitive impairment to dementia greatly varies between individuals, complicating prognosis and the assessment of much needed disease-modifying drugs. As a result there is a huge demand for greater understanding of the between-subject variability in progression and a need to understand biomarkers and risk factors for predicting future cognitive decline. In cohorts derived from two of the largest NHS foundation trust mental health service providers in the UK and multiple Alzheimer’s cohorts, this thesis explores methods for modelling disease progression and investigates the relationship between blood based proteins, genetic variants, health indicators and potential repurposing medications with cognitive decline. Through applying approaches that tackle three areas that require consideration for modelling of cognitive decline and disease progression, this thesis identified associations of antidepressant medication and psychotic symptoms associated with faster cognitive decline in dementia, in two separate cohorts.

Pharmacometric modelling is widely used in many aspects related to type 2 diabetes mellitus (T2DM), for instance in the anti-diabetes drug development, and in quantifying the disease progression of T2DM. The aim of this thesis were to improve the design of early phase anti-diabetes drug development studies with the focus on the power to identify mechanism of drug action (MoA), and to characterize and quantify the progression from prediabetes to overt diabetes, both the natural progression and the progression with diet and exercise interventions, using pharmacometrics modelling. The appropriateness of a study design depends on the MoAs of the anti-hyperglycaemic drug. Depending on if the focus is power to identify drug effect or accuracy and precision of drug effect, the best design will be different. Using insulin measurements on top of glucose has increase the power to identify a correct drug effect, distinguish a correct MoA from the incorrect, and to identify a secondary MoA in most cases. The accuracy and precision of drug parameter estimates, however, was not affected by insulin. A natural diabetes disease progression model was successfully added in a previously developed model to describe parameter changes of glucose and insulin regulation among impaired glucose tolerance (IGT) subjects, with the quantification of the lifestyle intervention. In this model, the assessment of multiple short-term provocations was combined to predict the long-term disease progression, and offers apart from the assessment of the onset of T2DM also the framework for how to perform similar analysis. Another previously published model was further developed to characterize the weight change in driving the changes in glucose homeostasis in subjects with IGT. This model includes the complex relationship between dropout from study and weight and glucose changes. This thesis has provided a first written guidance in designing a study for pharmacometrics analysis when characterizing drug effects, for early phase anti-diabetes drug development. The characterisation of the progression from prediabetes to overt diabetes using pharmacometrics modelling was successfully performed. Both the natural progression and the progression with diet and exercise interventions were quantified in this thesis.

This thesis was a step forward in gaining insight into falls in people with early stages of Parkinson's disease (PD), and in monitoring the disease progression based on clinical assessments. This research contributes new knowledge by providing new insights into utilizing information provided by the clinically administered instruments used routinely for the assessment of PD severity. The novel approach to modelling the progression of PD symptoms using multi-variable clinical assessment measurements for longitudinal data provides a new perspective into disease progression.

Type 2 diabetes mellitus (T2DM) is a metabolic disorder characterized by consistently high blood glucose, resulting from a combination of insulin resistance and reduced capacity of β-cells to secret insulin. While the exact causes of T2DM is yet unknown, obesity is known to be a major risk factor as well as co-morbidity for T2DM. As the global prevalence of obesity continues to increase, the association between obesity and T2DM warrants further study. Traditionally, mathematical models to study T2DM were mostly empirical and thus fail to capture the dynamic relationship between glucose and insulin. More recently, mechanism-based population models to describe glucose-insulin homeostasis with a physiological basis were proposed and offered a substantial improvement over existing empirical models in terms of predictive ability. The primary objectives of this thesis are (i) examining the predictive usefulness of semi-mechanistic models in T2DM by applying an existing population model to clinical data, and (ii) exploring the relationship between obesity and T2DM and describe it mathematically in a novel semi-mechanistic model to explain changes to the glucose-insulin homeostasis and disease progression of T2DM. Through the use of non-linear mixed effects modelling, the primary mechanism of action of an antidiabetic drug has been correctly identified using the integrated glucose-insulin model, reinforcing the predictive potential of semi-mechanistic models in T2DM. A novel semi-mechanistic model has been developed that incorporated a relationship between weight change and insulin sensitivity to describe glucose, insulin and glycated hemoglobin simultaneously in a clinical setting. This model was also successfully adapted in a pre-clinical setting and was able to describe the pathogenesis of T2DM in rats, transitioning from healthy to severely diabetic. This work has shown that a previously unutilized biomarker was found to be significant in affecting glucose homeostasis and disease progression in T2DM, and that pharmacometric models accounting for the effects of obesity in T2DM would offer a more complete physiological understanding of the disease.

This thesis examines the impact of HIV superinfection (infection of HIV-positive individuals by a heterologous HIV strain after immune responses have been established against the first strain) upon HIV disease progression and viral blips, and the relationship between non-adherence to cART and the occurrence of viral blips. For these purposes, a mathematical model of HIV within-host dynamics with two strains has been developed. My results suggest: firstly, HIV superinfection in and of itself was found not leading to faster progression to AIDS; it is only superinfection with strains of a higher replication capacity that does. Secondly, it was found that superinfecting strains susceptible to the existing cART regimen cannot establish themselves in patients, while those resistant to the regiment will lead to treatment failure. Superinfection in either scenario will not lead to viral blips. Thirdly, the choice of sampling frame was found to have a significant impact upon the observed number and incidence of viral blips. Instead of calculating the incidence of blips from their observed number over a period of time, one should take into account the sampling frame and calculate the proportion of blips among the measurements made over that period. Fourthly, increased drug adherence three days before a clinic visit does not mask poor adherence; regular consecutive non-adherence results in more blips than a random non-adherence pattern; and dose-timing variation around the regimen-prescribed time leads to more blips. Fifthly, the non-linear relationship between the proportion of measurements with detectable viral blips and the probable drug adherence of a patient, and how this relationship varies with the viral replication rate, are studied. This thesis improves our understanding of anti-HIV immune responses, refines our public health messages and provides us with indications of drug adherence through observation of viral blips with different sampling frames.

Le langage est une caractéristique déterminante de l'être humain et depuis des siècles les chercheurs s'intéressent à l'organisation fonctionnelle du langage et aux substrats neuronaux sous-tendent son fonctionnement normal. Le dysfonctionnement des mécanismes permettant un bon fonctionnement langagier est présent dans différentes maladies neurodégénératives, devenues ainsi un modèle majeur pour explorer les capacités langagières. En l'absence de traitement efficace pour les troubles du langage dans différentes maladies neurodégénératives, les techniques non invasives de stimulation cérébrale gagnent du terrain. Parmi ces techniques, la stimulation transcrânienne à courant continu (STCC) module l'activité neuronale via l'induction de faibles courants électriques dans le cerveau, et des effets bénéfiques ont été démontrés chez des patients aphasiques victimes d'accidents vasculaires cérébraux et des patients neurodégénératifs. Les études incluses dans cette thèse ont utilisé des modèles de lésions neurodégénératives pour étudier les mécanismes comportementaux de l'accès et du traitement des mots, pour étudier leur impact sur les capacités langagières et pour explorer la possibilité de moduler le langage à travers la STCC afin de définir sa valeur en tant qu'outil thérapeutique. Le manuscrit est divisé en 4 chapitres s'articulant autour de trois axes principaux : (1) recherche fondamentale sur le langage, (2) recherche clinique sur le dysfonctionnement du langage et approches thérapeutiques et (3) impact de facteurs individuels sur la variabilité de la réponse à ces thérapies, ainsi qu’un chapitre d’Introduction et un chapitre de Discussion Générale
Language is one of the most defining features of human beings and for centuries researchers have been interested on the functional organization of language and which neural substrates subtend its normal functioning. A breakdown of mechanisms subtending normal language abilities characterizes different neurodegenerative conditions, which have become models to study the neural basis and mechanisms of language processing. In the absence of effective treatments for language deficits in different neurodegenerative diseases, non-invasive brain stimulation approaches have been gaining momentum. Transcranial Direct Current Stimulation (tDCS) modulates neural activity via the induction of weak electrical intracranial currents, showing benefits in post-stroke and neurodegenerative aphasic patients. In this context, the studies included in this thesis analyzed neurodegenerative lesion models to characterize the behavioral mechanisms of word access and processing, address their impact on language abilities and explore the modulation of language impairment by means of tDCS to define its therapeutic value. The manuscript is divided in 4 chapters organized along three main axes: (1) fundamental research on language (2) clinical research on language breakdown and therapies and (3) impact of individual factors on the variability of the response to such therapies, an Introduction chapter and a General Discussion chapter

Pharmacodynamic processes and disease progression are increasingly characterized with pharmacometric models. However, modelling options for discrete-type responses remain limited, although these response variables are commonly encountered clinical endpoints. Types of data defined as discrete data are generally ordinal, e.g. symptom severity, count, i.e. event frequency, and time-to-event, i.e. event occurrence. Underlying assumptions accompanying discrete data models need investigation and possibly adaptations in order to expand their use. Moreover, because these models are highly non-linear, estimation with linearization-based maximum likelihood methods may be biased. The aim of this thesis was to explore pharmacometric methods and novel models for discrete data through (i) the investigation of benefits of treating discrete data with different modelling approaches, (ii) evaluations of the performance of several estimation methods for discrete models, and (iii) the development of novel models for the handling of complex discrete data recorded during (pre-)clinical studies. A simulation study indicated that approaches such as a truncated Poisson model and a logit-transformed continuous model were adequate for treating ordinal data ranked on a 0-10 scale. Features that handled serial correlation and underdispersion were developed for the models to subsequently fit real pain scores. The performance of nine estimation methods was studied for dose-response continuous models. Other types of serially correlated count models were studied for the analysis of overdispersed data represented by the number of epilepsy seizures per day. For these types of models, the commonly used Laplace estimation method presented a bias, whereas the adaptive Gaussian quadrature method did not. Count models were also compared to repeated time-to-event models when the exact time of gastroesophageal symptom occurrence was known. Two new model structures handling repeated time-to-categorical events, i.e. events with an ordinal severity aspect, were introduced. Laplace and two expectation-maximisation estimation methods were found to be performing well for frequent repeated time-to-event models. In conclusion, this thesis presents approaches, estimation methods, and diagnostics adapted for treating discrete data. Novel models and diagnostics were developed when lacking and applied to biological observations.

Les progrès de la recherche sur le cancer du sein dépendent de la disponibilité d’outils appropriés, comme les lignées cellulaires qui peuvent être implantées chez des souris immunocompétentes. La souche de souris C57Bl/6 est la plus étudiée et c’est la seule pour laquelle certaines variantes génétiques sont disponibles. Étant donné qu'aucune lignée cellulaire de carcinome mammaire à récepteurs hormonaux positifs de souche C57Bl/6 n'est disponible, nous avons décidé d'établir des lignées cellulaires de ce type. Nous avons induit des cancers du sein chez des souris C57BL/6 femelles en utilisant un analogue synthétique de la progestérone combiné à un agent endommageant l'ADN. Des lignées cellulaires ont été établies à partir de ces tumeurs et sélectionnées pour leur positivité au niveau du double récepteur (estrogène + progestérone), ainsi que pour leur transplantabilité chez les femelles C57BL/6. Parmi plusieurs lignées, une lignée cellulaire, que nous avons appelée MD5, remplissait ces critères et a permis l'établissement de tumeurs mal différenciées et très prolifératives. Ces tumeurs ont réduit leur croissance (sans toutefois régresser) lors du traitement par des antagonistes des récepteurs d’œstrogènes, ainsi que par une chimiothérapie à base d'anthracylines. Cependant, ce dernier effet n'a pas été influencé par la déplétion des lymphocytes T et, en outre, ces tumeurs n'ont pas répondu au blocage de PD-1, ce qui suggère que les tumeurs MD5 sont immunologiquement froides. En conclusion, les cellules MD5, dérivées des animaux C57BL/6, constituent un modèle de cancer du sein à récepteurs hormonaux positifs de mauvais pronostic
Progress in breast cancer research relies on the availability of suitable cell lines that can be implanted in immunocompetent laboratory mice. The best explored mouse strain, C57Bl/6, is also the only one for which multiple genetic variants are available. Driven by the fact that no hormone receptor-positive C57Bl/6-derived mammary carcinoma cell lines are available, we decided to establish such cell lines. Breast cancers were induced in female C57BL/6 mice using a synthetic progesterone analogue combined with a DNA damaging agent. Cell lines were established from these tumors and selected for dual (estrogen + progesterone) receptor positivity, as well as transplantability into C57BL/6 females. One cell line, which we called MD5,fulfilled these criteria and allowed for the establishment of poorly differentiated, highly proliferative, immune cold tumors. Such tumors reduced their growth (though did not regress) upon treatment with estrogen receptor antagonists, as well as with anthracyline-based chemotherapy. However, the latter effect was not influenced by T cell depletion and MD tumors failed to respond to PD-1 blockade, suggesting that they are immunologically cold. In conclusion, C57BL/6-derived MD5 cells constitute a model of poor prognosis hormone receptor-positive breast cancer

Tese de mestrado, Bioinformática e Biologia Computacional, Universidade de Lisboa, Faculdade de Ciências, 2021
Alzheimer’s Disease (AD) is a progressive brain disorder that slowly leads to memory loss, confusion, disorientation, and inability to communicate. It is very important an early detection of the disease in order to improve patients’ life quality and slow down the symptoms. Since there is still no cure available (although specific medications may attenuate the symptoms for a time), it ultimately draws from family members and society. Measuring and estimating the progression of such disease is therefore very important from both the medic and family’s perspective. Several studies have been made to address problems such as Alzheimer’s disease diagnosis and prognosis by discovering biomarkers. However, only a few studies use temporal information to model disease progression patterns. Hence, the objective of this thesis is to model the progression patterns of the dis ease through neuropsychological tests, leading to a better understanding of the underlying disease mechanisms and improving prognosis. In that sense, several summarization and representation techniques were applied to the dataset composed by neuropsychological tests, and the performance of classification techniques were assessed. Experimental results showed that representation techniques, such as ESiG, have a higher sensitivity and specificity values than others summarization and representation techniques or even the static data, using one medical appointment to predict the progression of Alzheimer’s disease.
A doença de Alzheimer é uma doença progressiva no cérebro que lentamente leva a perda de memória, confusão, desorientação e incapacidade de comunicar. É muito importante a deteção precoce da doença para ser possível aumentar a qualidade de vida a abrandar os sintomas. Atualmente ainda não existe uma cura para esta doença (apesar de medicação especifica pode atenuar os sintomas por um tempo) em última análise, muito desejado por membros da família e sociedade. Quantificar e estimar a progressão de uma doença como esta é muito importante na perspetiva médica e familiar. Diversos estudos foram realizados na abordagem de problemas como o diagnóstico da doença de Alzheimer e prognósticos por descoberta de biomarcadores. Contudo, apenas alguns estudos usam informação temporal para modelar os padrões de progressão da doença. Consequentemente, o objetivo desta tese é modelar padrões de progressão da doença com testes neuropsicológicos, para conduzir uma melhor compreensão dos mecanismos subjacentes da doença e melhorar o prognóstico. Neste Sentido, foram aplicadas diversas técnicas de sumarização e de representação ao conjunto de dados composto por testes neuropsicológicos, e avaliado o desempenho dos classificadores. Os resultados experimentais mostram que técnicas de representação, tais como o ESiG, apresentam valores de sensibilidade e de especificidade maiores que outras técnicas de summarização e de representação, ou mesmo utilizando valores de apenas uma consulta médica para prever a progressão para doença de Alzheimer.

The theme of this dissertation focuses on developing statistical models to learn progression dynamics and mechanisms of neuropsychiatric disorders using data from various domains. Due to limited knowledge about the underlying pathological processes in neurological disorders, it remains a challenge to establish reliable diagnostic criteria and predict disease prognosis in the presence of substantial phenotypic heterogeneity. As a result, current diagnosis and treatment of neurological disorders often rely on late-stage clinical symptoms, which poses barriers for developing effective interventions at the premanifest stage. It is crucial to characterize the temporal disease progression course and study the underlying mechanisms using clinical assessments, blood biomarkers, and neuroimaging biomarkers to evaluate disease stages, identify markers that are useful for early clinical diagnosis, compare or monitor treatment effects and accelerate drug discovery. We propose three projects to tackle challenges in leveraging multi-domain biomarkers and clinical symptoms to learn disease dynamics and progression of neurological disorders: (1) A nonlinear mixture model with subject-specific random inflection points to jointly fit multiple longitudinal markers and estimate marker progression trajectories in a single modality; (2) A multi-layer exponential family factor model integrating multi-domain data to learn lower-dimensional latent space of disease impairment and fully map disease risk and progression; (3) A latent state space model that jointly analyzes multi-channel EEG signals and learns dynamics of different sources corresponding to brain cortical activities. In addition, motivated by the ongoing COVID-19 pandemic, we propose a parsimonious survival-convolution model to predict daily new cases and estimate the time-varying reproduction numbers to evaluate effects of mitigation strategies. In the first project, we propose a nonlinear mixture model with random time shifts to jointly estimate long-term progression trajectories using multivariate discrete longitudinal outcomes. The model can identify early disease markers, their orders of occurrence, and the rates of impairment. Specifically, a latent binary variable representing disease susceptibility status incorporates subject covariates (e.g., biological measures) in the mixture model to capture between-subject heterogeneity. Measures of disease impairment for susceptible patients are modeled jointly under the exponential family framework. Our model allows for subject-specific and marker-specific inflection points associated with patients' characteristics (e.g., genetic mutation) to indicate a critical time when the fastest degeneration occurs. Furthermore, it uses subject-specific latent scores shared among markers to improve efficiency. The model is estimated using an EM algorithm. Extensive simulation studies are conducted to demonstrate validity of the proposed method and algorithm. Lastly, we apply our method to the Parkinson's Progression Markers Initiative (PPMI), and show utility to identify early disease signs and compare clinical symptomatology for the genetic form of Parkinson's Disease (PD) and idiopathic PD. In the second project, we tackle challenges to leverage multi-domain markers to learn early disease progression of neurological disorders. We propose to integrate heterogeneous types of measures from multiple domains (e.g., discrete clinical symptoms, ordinal cognitive markers, continuous neuroimaging and blood biomarkers) using a hierarchical Multi-layer Exponential Family Factor (MEFF) model, where the observations follow exponential family distributions with lower-dimensional latent factors. The latent factors are decomposed into shared factors across multiple domains and domain-specific factors, where the shared factors provide robust information to perform behavioral phenotyping and partition patients into clinically meaningful and biologically homogeneous subgroups. Domain-specific factors capture the remaining unique variations for each domain. The MEFF model also captures the nonlinear trajectory of disease progression and order critical events of neurodegeneration measured by each marker. To overcome computational challenges, we fit our model by approximate inference techniques for large-scale data. We apply the developed method to Parkinson's Progression Markers Initiative (PPMI) data to integrate biological, clinical and cognitive markers arising from heterogeneous distributions. The model learns lower-dimensional representations of Parkinson's disease and the temporal ordering of the neurodegeneration of PD. In the third project, we propose methods that can be used to analyze multi-channel electroencephalogram (EEG) signals intensively measured at a high temporal resolution. Modern neuroimaging technologies have substantially advanced the measurement of brain activities. EEG as a non-invasive neuroimaging technique measures changes in electrical voltage on the scalp induced by cortical activities. With its high temporal resolution, EEG has emerged as an increasingly useful tool to study brain connectivity. Challenges with modeling EEG signals of complex brain activities include interactions among unknown sources, low signal-to-noise ratio and substantial between-subject heterogeneity. In this work, we propose a state space model that jointly analyzes multi-channel EEG signals and learns dynamics of different sources corresponding to brain cortical activities. Our model borrows strength from spatially correlated measurements and uses low-dimensional latent sources to explain all observed channels. The model can account for patient heterogeneity and quantify the effect of a subject's covariates on the latent space. The EM algorithm, Kalman filtering, and bootstrap resampling are used to fit the state space model and provide comparisons between patient diagnostic groups. We apply the developed approach to a case-control study of alcoholism and reveal significant attenuation of brain activities in response to visual stimuli in alcoholic subjects compared to healthy controls. Lastly, motivated by the ongoing COVID-19 pandemic, we propose a robust and parsimonious survival-convolution model aiming to predict COVID-19 disease course and compare effectiveness of mitigation measures across countries to inform policy decision making. We account for transmission during a pre-symptomatic incubation period and use a time-varying effective reproduction number to reflect the temporal trend of transmission and change in response to a public health intervention. We estimate the intervention effect on reducing the infection rate using a natural experiment design and quantify uncertainty by permutation. In China and South Korea, we predicted the entire disease epidemic using only early phase data (two to three weeks after the outbreak). A fast rate of decline in reproduction number was observed and adopting mitigation strategies early in the epidemic was effective in reducing the infection rate in these two countries. The nationwide lockdown in Italy did not accelerate the speed at which the infection rate decreases. In the United States, the reproduction number significantly decreased during a 2-week period after the declaration of national emergency, but declines at a much slower rate afterwards. If the trend continues after May 1, COVID-19 may be controlled by late July. However, a loss of temporal effect (e.g., due to relaxing mitigation measures after May 1) could lead to a long delay in controlling the epidemic.

abstract: Recently, a well-designed and well-trained neural network can yield state-of-the-art results across many domains, including data mining, computer vision, and medical image analysis. But progress has been limited for tasks where labels are difficult or impossible to obtain. This reliance on exhaustive labeling is a critical limitation in the rapid deployment of neural networks. Besides, the current research scales poorly to a large number of unseen concepts and is passively spoon-fed with data and supervision. To overcome the above data scarcity and generalization issues, in my dissertation, I first propose two unsupervised conventional machine learning algorithms, hyperbolic stochastic coding, and multi-resemble multi-target low-rank coding, to solve the incomplete data and missing label problem. I further introduce a deep multi-domain adaptation network to leverage the power of deep learning by transferring the rich knowledge from a large-amount labeled source dataset. I also invent a novel time-sequence dynamically hierarchical network that adaptively simplifies the network to cope with the scarce data. To learn a large number of unseen concepts, lifelong machine learning enjoys many advantages, including abstracting knowledge from prior learning and using the experience to help future learning, regardless of how much data is currently available. Incorporating this capability and making it versatile, I propose deep multi-task weight consolidation to accumulate knowledge continuously and significantly reduce data requirements in a variety of domains. Inspired by the recent breakthroughs in automatically learning suitable neural network architectures (AutoML), I develop a nonexpansive AutoML framework to train an online model without the abundance of labeled data. This work automatically expands the network to increase model capability when necessary, then compresses the model to maintain the model efficiency. In my current ongoing work, I propose an alternative method of supervised learning that does not require direct labels. This could utilize various supervision from an image/object as a target value for supervising the target tasks without labels, and it turns out to be surprisingly effective. The proposed method only requires few-shot labeled data to train, and can self-supervised learn the information it needs and generalize to datasets not seen during training.
Dissertation/Thesis
Doctoral Dissertation Computer Science 2020

Indiana University-Purdue University Indianapolis (IUPUI)
Although rare, classification of over 200 hereditary cancer susceptibility syndromes accounting for ~5-10% of cancer incidence has enabled the discovery and understanding of cancer predisposition genes that are also frequently mutated in sporadic cancers. The need to prevent or delay invasive cancer can partly be addressed by characterization of cells derived from healthy individuals predisposed to cancer due to inherited "single-hits" in genes in order to develop patient-derived samples as preclinical models for mechanistic in vitro studies. Here, we present microarray-based transcriptome profiling of Li-Fraumeni syndrome (LFS) patient-derived unaffected breast epithelial cells and their phenotypic characterization as in vitro three-dimensional (3D) models to test pharmacological agents. In this study, the epithelial cells derived from the unaffected breast tissue of a LFS patient were cultured and progressed from non-neoplastic to a malignant stage by successive immortalization and transformation steps followed by growth in athymic mice. These cell lines exhibited distinct transcriptomic profiles and were readily distinguishable based upon their gene expression patterns, growth characteristics in monolayer and in vitro 3D cultures. Transcriptional changes in the epithelial-to-mesenchymal transition gene signature contributed to the unique phenotypes observed in 3D culture for each cell line of the progression series; the fully transformed LFS cells exhibited invasive processes in 3D culture with disorganized morphologies due to cell-cell miscommunication, as seen in breast cancer. Bioinformatics analysis of the deregulated genes and pathways showed inherent differences between these cell lines and targets for pharmacological agents. After treatment with small molecule APR-246 that restores normal function to mutant p53, we observed that the neoplastic LFS cells had reduced malignant invasive structure formation from 73% to 9%, as well as an observance of an increase in formation of well-organized structures in 3D culture (from 27% to 91%) by stereomicroscopy and confocal microscopy. Therefore, the use of well-characterized and physiologically relevant preclinical models in conjunction with transcriptomic profiling of high-risk patient derived samples as a renewable laboratory resource can potentially guide the development of safer and more effective chemopreventive approaches.