Détection Pan Génomique de Locus Sous Sélection en Présence de Données Multi-populationnelles en Marquage Dense : Nouvelles Méthodes Multipoint Et Applications Aux Espèces Animales D'élevage

Since initial domestication by Humans, farm animal species have experienced great phenotype diversification and thus represent an interesting model for the study of natural and artificial selection. Besides, the detection of selection signatures in these species can have substantial agronomic outcomes, pointing out genomic regions related to production traits or resistance to diseases. Many datasets giving access to genome wide genotypic information have become available enabling to scan entire genomes for signatures of selection. Many tests designed to detect positive selection are challenged by at least one of the following problems. First, as the number of available markers increases, so do their correlations which need to be taken into account. Second, many tests have been designed to compare populations only pairwise. Considering more than two populations simultaneously should increase the power to detect selected regions, but necessitates to account for correlations between them arising from their shared history. I proposed two statistical tests for the detection of positive selection signatures using dense genetic data collected from multiple populations. One is based on haplotypic differentiation between populations, requiring genetic data at the individual level. The second one consists in cumulating single marker signals using local score theory, requiring data at the population level. Through simulated and real datasets, I showed that these tests increased the detection power compared to other existing tests in many cases. Applied to two data sets in sheep and quail, they also point out biologically relevant candidate genes under selection.