Statistical tools for genome-wide studies

The aim of genomic selection (GS) in livestock is to detect linkage disequilibrium between SNP and quantitative trait loci (QTL) across the whole genome, to improve the accuracy of the estimated breeding value (GEBV) in genetic improvement programs. Two main issues affect GS: the imbalance between the number of SNP and the number of involved animals and the high genotyping costs.In this thesis the principal component analysis (PCA) is proposed as a method to reduce the dimensionality of the SNP data. In particular, the study evaluated the effect of the rank of the variance-covariance matrix on the accuracy of GEBV when PCA was applied.In addition, a new approach is proposed to reduce the dimensionality of the data. First, this new method was used in a genomic wide association study to detect associations among markers and traits under study. Then the obtained results were used to reduce the number of SNPs useful to estimate the GEBV. Results show that, the accuracy of GEBV, when only the SNPs selected with the new method were used, was on average nearly equal to or sometimes greater than the accuracies obtained when all SNPs were used.This thesis also proposes the partial least squared regression (PLSR) to impute markers not present in economic chips and avoid a reduction in the accuracy of GEBV estimation. The study demonstrated that the PLSR imputation method can efficiently impute missing genotypes from low-density panels to HDP.