Files
Abstract
The availability of large-scale genotyping platforms provided an unprecedented resource to dissect associations between traits of interest and genomic variation, and to enhance the estimation of breeding through genomic selection (GS) for animal and plant applications. The dramatic increase in the number of variants was expected to significantly increase the accuracy of GS. Unfortunately, that was not the case due to several factors including the huge increase in the number of variants in the association model, increase in co-linearity, and the reduction in statistical power. Thus, accuracy of GS using either multiple regression (RM) or mixed linear models (VC) approaches did not result in any significant increase in the accuracy of GS after a certain density of SNP markers in the genotyping panels was reached. It was clear that SNP prioritization is needed in order to harness the full potential of high density marker panels. Wrights fixation index (FST), a measure the genetic differentiation among sub-populations, was proposed to prioritize SNPs in high-density marker panels and to track relevant quantitative trait loci (QTL). Using the phenotypic distribution, three sub-populations were created based on the 5 and 95% quantiles and were used to identify markers under selection pressure. Genomic data consisted of 200K and 400K SNP markers distributed on 10 chromosomes to mimic 770K and 1.2 million SNPs markers in the bovine genome. In the first study, the performance of FST prioritized SNPs was compared to RM based methods. Only 0.5 to 1% of SNPs were needed to achieve the same accuracy as used all markers in the panel. Furthermore, using FST method showed slight superiority compared to BayesB and BayesC. In the second study, the effects of FST prioritized SNPs on the computation of the genomic relationship matrix (G), genomic similarity and accuracy of GS were assessed. The results showed that a balance between genomic similarity and percentage of genetic variance explained is needed to optimize the accuracy of GS. In the third paper, FST scores were used to derive weights for computing G. The results clearly showed that accuracy could be improved under different weighting scenarios.