High-dimensional statistical models for a better genomic evaluation of quantitative traits
Animal breeding aims at the improvement of performance and health traits by exploiting the heritable impact on a trait. Genomic markers (e.g. SNPs) are used to assess the genetic variation in a population. We develop mathematical-statistical methods to investigate the relationship between genetic variation and trait characteristic and to estimate the size of genetic impact. It is a challenge to identify among the multitude of genomic markers those that are really relevant for a trait. We achieve this by applying special shrinkage and selection methods.
Dependencies between SNPs occur due to linkage and linkage disequilibrium among chromosome segments, and the population structure affects the extent of such dependencies. A typical livestock population has a certain family stratification; we investigate half-sib (e.g. dairy cattle) and full-sib families (e.g. chicken). We determine the dependence between SNP pairs by estimating recombination rate and linkage disequilibrium from SNP genotypes. This additional information is then taken into account in the genomic evaluation of quantitative traits. We also investigate how genomic markers can be meaningfully grouped according to their interdependence in order to reduce the statistical model.