No description available.

In healthcare, vulnerable populations that are using medications with a narrow therapeutic index and wide interpatient PK/PD (pharmacokinetic/pharmacodynamic modelling) variability are increasing. As such, variable dosage regimens may result in severe therapeutic failures or adverse drug reactions (ADR). Improved monitoring of patient response to medication and personalization of treatment is therefore warranted. Precision dosing aims to individualize drug regimens for each patient based on independent factors obtained from a patient’s clinical records. Personalization of dosing increases the accuracy and efficiency of medication delivery. This can be achieved through utilizing the wide range of Electronic Health Records (EHR) contain the patients’ medical history, diagnoses, laboratory test results, demographics, treatment plans, biomarker data; information that can be exploited to generate a patient-specific treatment regimen. For example, Fast Healthcare Interoperability Resources (FHIR) is an existing healthcare standard that provides a framework on which semantic exchange of meaningful clinical information can be developed such as using an ontology as a decision support tool to achieve precision medicine. The purpose of this thesis is to make an investigation of the feasibility of interoperability in EHR and propose an ontology framework for precision dosing using currently existing health standards. The methodology involved carrying out of semi-structured interviews from professionals in relevant areas of expertise and document analysis of already existent literature, a precision dosing ontology framework is developed. Results show key tenants for an ontology framework and drugs and their covariates. The thesis therefore advances to investigate how data requirements in EHR systems, IT platforms, implementation, and integration of Model Imposed Precision Dosing (MIPD) and recommendations have been evaluated to cater to interoperability. With modern healthcare striving for personalized healthcare, precision medicine would offer an improved therapeutic experience for a patient.

La dégénérescence maculaire liée à l’âge (DMLA) est la première cause de cécité dans les pays industrialisés. C’est une maladie complexe et multifactorielle ayant des conséquences majeures sur la qualité de vie des personnes atteintes. De nombreux facteurs de risque, génétiques et non génétiques, jouent un rôle important dans la pathogénèse des stades avancés de la DMLA. Les modèles de prédiction développés à ce jour reposent sur un nombre limité de ces facteurs, et sont encore peu utilisés dans la pratique clinique.Ce travail de thèse avait pour premier objectif d’identifier de nouveaux biomarqueurs circulants du risque de DMLA. Ainsi, à partir d’une étude post-mortem basée sur une approche de lipidomique, nous avons identifié les composés lipidiques sanguins les plus prédictifs des concentrations rétiniennes en acides gras polyinsaturés omégas 3 (AGPI w-3). Nous avons développé un modèle de prédiction basé sur 7 espèces de lipides des esters de cholestérol. Ce modèle, obtenu en combinant pénalisation et réduction de la dimension, a ensuite été validé dans des études cas-témoins de DMLA et dans un essai clinique randomisé de supplémentation en AGPI w-3. Ces biomarqueurs pourraient être utiles pour l’identification des personnes à haut risque de DMLA, qui pourraient ainsi bénéficier d’une supplémentation en AGPI w-3.Le deuxième objectif de cette thèse était de développer un modèle de prédiction du risque de progression vers une DMLA avancée à partir de facteurs de risque génétiques, phénotypiques et environnementaux. Une originalité de notre travail a été d’utiliser une méthode de régression pénalisée – un algorithme d’apprentissage automatique – dans un cadre de survie afin de tenir compte de la multicollinéarité entre les facteurs de risque. Nous avons également pris en compte la censure par intervalle et le risque compétitif du décès via un modèle à 3 états sain-malade-mort. Nous avons ensuite validé ce modèle sur une étude indépendante en population générale.Il serait intéressant de valider ce modèle de prédiction dans d’autres études indépendantes en y incluant les biomarqueurs circulants identifiés à partir de l’étude de lipidomique effectuée dans le cadre de cette thèse. Le but final serait d’intégrer cet outil prédictif dans la pratique clinique afin de rendre la médecine de précision une réalité pour les patients atteints de DMLA dans le futur proche. / Age-related macular degeneration (AMD) is the leading cause of blindness in industrialized countries. AMD is a complex and multifactorial disease with major consequences on the quality of life. Numerous genetic and non-genetic risk factors play an important role in the pathogenesis of the advanced stages of AMD. Existing prediction models rely on a restricted set of risk factors and are still not widely used in the clinical routine.The first objective of this work was to identify new circulating biomarkers of AMD’s risk using a lipidomics approach. Based on a post-mortem study, we identified the most predictive circulating lipids of retinal content in omega-3 polyunsaturated fatty acids (w-3 PUFAs). We combined penalization and dimension reduction to establish a prediction model based on plasma concentration of 7 cholesteryl ester species. We further validated this model on case-control and interventional studies. These biomarkers could help identify individuals at high risk of AMD who could be supplemented with w-3 PUFAs.The second objective of this thesis was to develop a prediction model for advanced AMD. This model incorporated a wide set of phenotypic, genotypic and lifestyle risk factors. An originality of our work was to use a penalized regression method – a machine learning algorithm – in a survival framework to handle multicollinearities among the risk factors. We also accounted for interval censoring and the competing risk of death by using an illness-death model. Our model was then validated on an independent population-based cohort.It would be interesting to integrate the circulating biomarkers identified in the lipidomics study to our prediction model and to further validate it on other external cohorts. This prediction model can be used for patient selection in clinical trials to increase their efficiency and paves the way towards making precision medicine for AMD patients a reality in the near future.

<p>Graphs are a natural choice for understanding the associations between variables, and assuming a probabilistic embedding for the graph structure leads to a variety of graphical models that enable us to understand these associations even further. In the realm of high-dimensional data, where the number of associations between interacting variables is far greater than the available number of data points, the goal is to infer a sparse graph. In this thesis, we make contributions in the domain of Bayesian graphical models, where our prior belief on the graph structure, encoded via uncertainty on the model parameters, enables the estimation of sparse graphs.</p> <p><br></p> <p>We begin with the Gaussian Graphical Model (GGM) in Chapter 2, one of the simplest and most famous graphical models, where the joint distribution of interacting variables is assumed to be Gaussian. In GGMs, the conditional independence among variables is encoded in the inverse of the covariance matrix, also known as the precision matrix. Under a Bayesian framework, we propose a novel prior--penalty dual called the `graphical horseshoe-like' prior and penalty, to estimate precision matrix. We also establish the posterior convergence of the precision matrix estimate and the frequentist consistency of the maximum a posteriori (MAP) estimator.</p> <p><br></p> <p>In Chapter 3, we develop a general framework based on local linear approximation for MAP estimation of the precision matrix in GGMs. This general framework holds true for any graphical prior, where the element-wise priors can be written as a Laplace scale mixture. As an application of the framework, we perform MAP estimation of the precision matrix under the graphical horseshoe penalty.</p> <p><br></p> <p>In Chapter 4, we focus on graphical models where the joint distribution of interacting variables cannot be assumed Gaussian. Motivated by the quantile graphical models, where the Gaussian likelihood assumption is relaxed, we draw inspiration from the domain of precision medicine, where personalized inference is crucial to tailor individual-specific treatment plans. With an aim to infer Directed Acyclic Graphs (DAGs), we propose a novel quantile DAG learning framework, where the DAGs depend on individual-specific covariates, making personalized inference possible. We demonstrate the potential of this framework in the regime of precision medicine by applying it to infer protein-protein interaction networks in Lung adenocarcinoma and Lung squamous cell carcinoma.</p> <p><br></p> <p>Finally, we conclude this thesis in Chapter 5, by developing a novel framework to compute the marginal likelihood in a GGM, addressing a longstanding open problem. Under this framework, we can compute the marginal likelihood for a broad class of priors on the precision matrix, where the element-wise priors on the diagonal entries can be written as gamma or scale mixtures of gamma random variables and those on the off-diagonal terms can be represented as normal or scale mixtures of normal. This result paves new roads for model selection using Bayes factors and tuning of prior hyper-parameters.</p>

Les leucémies myéloïdes aiguës (LMA) consistent en un groupe de cancers agressifs causés par une accumulation de mutations génétiques et épigénétiques survenant dans les cellules souches ou progénitrices de la moelle osseuse. Il s’agit d’un groupe de maladies très hétérogène, caractérisé par un grand nombre de combinaisons d’altérations qui perturbent à la fois les voies de signalisation qui y sont exprimées, leur sensibilité aux différents traitements et le pronostic des patients. Le déploiement des technologies de séquençage de nouvelle génération au courant de la dernière décennie a permis l’exploration à une échelle sans précédent du paysage mutationnel et transcriptomique de différents cancers, incluant les LMA. Dans le cadre de nos travaux, nous avons voulu tester l'hypothèse selon laquelle les LMA se déclinent en plusieurs sous-groupes génétiques caractérisés chacun par des mutations distinctes et une expression génique dérégulée, ainsi qu’une réponse différentielle à des molécules qui pourraient représenter de nouvelles stratégies thérapeutiques. Nous avons testé cette hypothèse au sein de la cohorte Leucegene, qui comprend un grand nombre de LMA primaires analysées par le séquençage du transcriptome, et nous avons analysé les différences entre les différents sous-groupes en les analysant un à la fois. Cette étude des différents sous-groupes nous a permis de disséquer le profil génomique, transcriptomique et les sensibilités aux petites molécules de sept sous-groupes génétiques, représentant environ la moitié des cas de LMA de l’adulte. Notre approche a permis de découvrir plusieurs nouvelles mutations spécifiques aux différents sous-groupes, dont certaines ont été validées dans des cohortes indépendantes. Nous avons également confirmé que les gènes différentiellement exprimés dans les sous-groupes sont plus informatifs que les signatures d'expression non supervisées pour identifier les biomarqueurs de la maladie. Nous avons ainsi identifié dans la majorité des sous-groupes des gènes représentant un biomarqueur d'intérêt, ayant une pertinence fonctionnelle ou pronostique. Ces données ont également mené à des criblages chimiques ciblés qui ont identifié de nouvelles vulnérabilités dépendant du contexte génétique. Au-delà de ces observations, nos travaux pourraient avoir une portée translationnelle tandis que le séquençage de nouvelle génération est de plus en plus utilisé en clinique. La combinaison avec d’autres modalités de séquençage et l’incorporation de technologies émergentes aideront à poursuivre la dissection génomique, transcriptomique et chimique de la LMA et l’approche utilisée pourra même éventuellement s’appliquer à d’autres types de cancers. / Acute myeloid leukemias (AML) are a group of cancers caused by an accumulation of genetic and epigenetic mutations occurring in the stem or progenitor cells of the bone marrow. They represent a very heterogeneous group of diseases, characterized by a large number of combinations of alterations which disrupt to varying degrees key networks in these cells, their sensitivity to treatments and the prognosis of the patients. The deployment of next-generation sequencing technologies over the past decade has enabled exploration on an unprecedented scale of the mutational and transcriptomic landscape of various cancers, including AML. As part of our work, we tested the hypothesis according to which AMLs comprise several genetic subgroups, each characterized by distinct mutations and deregulated gene expression profiles, as well as a differential response to molecules that could represent novel therapies. We tested this hypothesis in the Leucegene cohort, which includes a large number of primary AMLs analyzed by transcriptome sequencing, which we explored one subgroup after the other, dissecting the genomic, transcriptomic or small molecule sensitivities profile of seven AML subgroups representing approximately half of adult AML cases. Our approach has allowed us to discover several new mutations specific to different subgroups, some of which have been validated in independent cohorts. We also confirmed that genes differentially expressed in subgroups are more informative than unsupervised expression signatures, and we identified genes representing potential biomarkers, or having a functional or prognostic relevance in the majority of subgroups. Generated data also led to targeted chemical screens performed on primary AML cells, which identified new context-dependent vulnerabilities. Beyond these observations, our work could have a translational scope while next-generation sequencing is paving its way in the clinic. The combination with other Omics and the incorporation of emerging technologies will help to further the multi-dimensional dissection of these groups and additional ones, as the presented approach could be applied to additional disease subsets and cancer types.

Patients not responding to therapy as expected is one of the most pressing healthcare concerns of today. It causes economical, medical and societal issues along with suffering for patients. This project aimed to address this problem and evaluate how to find the best suited drug treatments for individual patients to treat Covid-19. This project was carried out in collaboration with the company AB Mavatar, that have two networks, one experimental and one predicted, which produce drug treatment rankings differently. Different methods are used to connect drug targets to disease associated genes and thus evaluate what drugs are best suited for specific patients to treat Covid-19. The aim of this project is to examine how network, method and drug category affect the ranking of a drug treatment for four mapped Covid-19 patients. Which drug category a drug belongs to did not seem to significantly affect the drug ranking. Yet, certain drug subcategories were closely correlated. However, these subcategories were not those that are typically associated with Covid-19. The method used to connect drug targets to disease associated genes heavily impacts the ranking of the drug treatment. The methods should be further evaluated to see if some should be excluded or weighted less in drug ranking calculations. The two networks are similar in how they rank different drugs, especially in severely ill patients. Through this project and the evaluation of the impact of method choice, one can start to figure out what should be prioritized among disease related changes. Also, important parameters for personalized treatment can be evaluated. / Patienter som inte svarar på terapi som förväntat är en av de största utmaningarna inom hälso- och sjukvård idag. Det orsakar ekonomiska, medicinska och samhälleliga problem samt lidande för patienter. Det här projektet adresserade detta problem och evaluerade hur man kan hitta det bäst lämpade läkemedlet för specifika patienter för att behandla Covid-19. Projektet gjordes tillsammans med företaget AB Mavatar, som har två interaktom, en experimentell och en datadriven, som rangordnar läkemedelsrekommendationer på olika sätt. Olika metoder används för att koppla samman läkemedelsmål med sjukdomsrelaterade gener och således evaluera vilka läkemedel som är bäst lämpade för specifika patienter för behandling av Covid-19. Syftet med projektet var att undersöka hur nätverk, metod och läkemedelskategori påverkar hur läkemedel rangordnas för fyra kartlagda Covid-19-patienter.  Vilken läkemedelskategori ett läkemedel tillhör tycks inte märkbart påverka läkemedelsrangordning. Trots detta var vissa läkemedelsunderkategorier nära korrelerade. Dock var dessa underkategorier inte typiskt associerade med Covid-19. Metoden för att koppla samman läkemedelsmål med sjukdomsassocierade gener påverkade läkemedelsrangordningen väsentligt. Metoderna borde dock evalueras ytterligare för att eventuellt exkludera eller vikta vissa mindre i uträkningar av läkemedelsrang. De två nätverken är lika i hur de rangordnar olika läkemedel, särskilt för svårt sjuka patienter. Genom detta projekt och genom evaluering av metodvalets påverkan kan man börja begripa hur man borde priorita bland sjukdomsrelaterade förändringar. Dessutom kunde viktiga parametrar inom personlig behandling evalueras.

Whole-exome and genome sequencing are widely used to diagnose individual patients. However, despite its success, this approach leaves many patients undiagnosed. This could be due to the need to discover more disease genes and variants or because disease phenotypes are novel and arise from a combination of variants of multiple known genes related to the disease. Recent rapid increases in available genomic, biomedical, and phenotypic data enable computational analyses, reducing the search space for disease-causing genes or variants and facilitating the prediction of causal variants. Therefore, artificial intelligence, data mining, machine learning, and deep learning are essential tools that have been used to identify biological interactions, including protein-protein interactions, gene-disease predictions, and variant--disease associations. Predicting these biological associations is a critical step in diagnosing patients with rare or complex diseases. In recent years, computational methods have emerged to improve gene-disease prioritization by incorporating phenotype information. These methods evaluate a patient's phenotype against a database of gene-phenotype associations to identify the closest match. However, inadequate knowledge of phenotypes linked with specific genes in humans and model organisms limits the effectiveness of the prediction. Information about gene product functions and anatomical locations of gene expression is accessible for many genes and can be associated with phenotypes through ontologies and machine-learning models. Incorporating this information can enhance gene-disease prioritization methods and more accurately identify potential disease-causing genes. This dissertation aims to address key limitations in gene-disease prediction and variant prioritization by developing computational methods that systematically relate human phenotypes that arise as a consequence of the loss or change of gene function to gene functions and anatomical and cellular locations of activity. To achieve this objective, this work focuses on crucial problems in the causative variant prioritization pipeline and presents novel computational methods that significantly improve prediction performance by leveraging large background knowledge data and integrating multiple techniques. Therefore, this dissertation presents novel approaches that utilize graph-based machine-learning techniques to leverage biomedical ontologies and linked biological data as background knowledge graphs. The methods employ representation learning with knowledge graphs and introduce generic models that address computational problems in gene-disease associations and variant prioritization. I demonstrate that my approach is capable of compensating for incomplete information in public databases and efficiently integrating with other biomedical data for similar prediction tasks. Moreover, my methods outperform other relevant approaches that rely on manually crafted features and laborious pre-processing. I systematically evaluate our methods and illustrate their potential applications for data analytics in biomedicine. Finally, I demonstrate how our prediction tools can be used in the clinic to assist geneticists in decision-making. In summary, this dissertation contributes to the development of more effective methods for predicting disease-causing variants and advancing precision medicine.