The fundamental theme my thesis explores is the relationship between genetic variation and phenotypic variation. It addresses three main questions. What is the genetic architecture of traits in the HS? How can sequence information help identifying the sequence variants and genes responsible for phenotypic variation? Are the genetic factors contributing to phenotypic variation in the rat homologous to those contributing to variation in the same phenotype in the mouse? To address these questions, I analysed data collected by the EURATRANS consortium on 1,407 Heterogeneous Stock (HS) rats descended from eight inbred strains through sixty generations of outbreeding. The HS rats were genotyped at 803,485 SNPs and 160 measures relevant to a number of models of disease (e.g. anxiety, type 2 diabetes, multiple sclerosis) were collected. The eight founders of the Stock were genotyped and sequenced. I identified loci in the genome that contribute to phenotypic variation (Quantitative Trait Loci, QTLs), and integrated sequence information with the mapping results to identify the genetic variants underlying the QTLs. I made some important observations about the nature of genetic architecture in rats, and how this compares to mice and humans. I also showed how sequence information can be used to improve mapping resolution, and in some cases to identify causal variants. However, I report an unexpected observation: at the majority of QTLs, the genetic effect cannot be accounted for by a single variant. This finding suggests that genetic variation cannot be reduced to sequence variation. This complexity will need to be taken into account by studies that aim at unravelling the genetic basis of complex traits.

Les aptitudes du lait à la transformation en fromage sont étroitement liées à sa composition, notamment en protéines. Ces caractères, difficiles à mesurer directement, ont été prédits à partir des spectres moyen infrarouge (MIR) du lait pour la composition en protéines dans les 3 races bovines Montbéliarde, Normande et Holstein (projet PhénoFinlait) et pour 9 aptitudes fromagères et la composition fine du lait en race Montbéliarde (projet From’MIR). La méthode Partial Least Squares (PLS) a fourni des prédictions MIR plus précises que les méthodes bayésiennes testées.Une analyse génétique a été réalisée pour ces caractères prédits à partir de plus de six millions de spectres MIR de plus de 400 000 vaches.Les caractères fromagers et de composition du lait sont modérément à fortement héritables. Les corrélations génétiques entre caractères fromagers (rendements et coagulation) et avec la composition du lait (protéines, acides gras et minéraux) sont élevées et favorables.Les génotypes de 28 000 vaches ont été imputés jusqu’à la séquence complète grâce aux données du projet 1000 génomes bovins.Des analyses d’association (GWAS) révèlent de nombreux gènes et variants avec des effets forts sur la fromageabilité et la composition du lait. Un réseau de 736 gènes, par ailleurs associé à ces caractères, permet d’identifier des voies métaboliques et des gènes régulateurs fonctionnellement liés à ces caractères.Un prototype d’évaluation génomique a été mis en place en race Montbéliarde. Un modèle de type contrôles élémentaires, incluant les variants détectés par les GWAS et présumés causaux, donne les estimations des valeurs génomiques les plus précises. La simulation d’une sélection incluant les caractères fromagers montre qu’il est possible d’améliorer la fromageabilité du lait avec un impact limité sur le gain génétique des autres caractères sélectionnés.Les travaux présentés dans cette thèse ont abouti 1) à la détection de gènes (dont certains jamais décrits auparavant) et de variants candidats pour la composition et la fromageabilité du lait et 2) à la mise en place d’une évaluation génomique de la fromageabilité du lait en race Montbéliarde dans la zone AOP Comté. / The ability of milk to be processed into cheese is closely linked to its composition, in particular in proteins. These traits, which are difficult to measure directly, were predicted from milk mid-infrared (MIR) spectra for protein composition in 3 cattle breeds Montbéliarde, Normande and Holstein (PhénoFinlait project) and for 9 milk cheese-making properties (CMP) and composition traits in Montbéliarde cows (From’MIR project). The Partial Least Squares method provided more accurate predictions than the Bayesian methods tested.A genetic analysis was performed on these traits, predicted from more than six million MIR spectra of more than 400,000 cows.Milk CMP and composition traits are moderately to highly heritable. Genetic correlations between CMP (cheese yields and coagulation) and with milk composition (proteins, fatty acids and minerals) are high and favorable.The genotypes of 28,000 cows were imputed to whole genome sequences using the 1000 bovine genome reference population.Genome wide association studies (GWAS) reveal many genes and variants in these genes with strong effects on CMP and milk composition. A network of 736 genes, associated with these traits, enable the identification of metabolic pathways and regulatory genes functionally linked to these traits.A pilot genomic evaluation was set up in Montbéliarde cows. A test-day model, including variants detected by GWAS, provides the most accurate genomic value estimates. Simulation of a selection shows that it is possible to improve the cheesability of milk with a limited impact on the genetic gain of the traits that currently make up the breeding objective.The work presented in this thesis led to 1) the detection of genes (some of which have never been described before) and candidate variants for milk CMP and composition traits and 2) the implementation of a genomic evaluation of CMP predicted from MIR spectra in Montbéliarde cows of the Comté PDO area.

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.