Efficiency of single-step GBLUP in genomic evaluation and GWAS in broiler chickens
MetadataShow full item record
Genomic selection has been a hot topic in the poultry industry during the last couple of years. Many tools have been built to conduct genomic evaluation and to inspect changes in genetic prediction before and after selection. Here, we evaluate selected features of the single-step genomic best linear unbiased prediction (ssGBLUP) statistical method. This method can predict genomic estimated breeding values (GEBVs) by blending traditional pedigree relationships with realized relationships derived from genetic markers. Subsequently, GEBVs can be utilized in a genome-wide association study (GWAS) by conversion of GEBVs to marker effects and their weights. The dissertation utilized ssGBLUP in 4 studies. In the first study, the signatures of selection in male and female broiler breeds selected for the same goals were analyzed. Results indicated that the male breed had undergone stronger selection compared with the female breed in terms of allele frequency change. Furthermore, female breed had a greater heterozygosity change compared with the male breed. No overlapping selection region was found in the two breeds. In the second study, five options for weighted ssGBLUP (WssGBLUP) were tested. Simulated data sets included 5, 100, and 500 quantitative trait loci (QTLs). Weights were calculated based on formulas for single or segment single-nucleotide polymorphism (SNP) variance. Prediction accuracy for WssGBLUP improved at 2nd to 4th iterations by updating the mean, max or summation of ui2 among every 20 (SNP), where ui is the effect of SNP i. Accuracy reached a plateau after iteration 3 or 5 by using weights proportional to ui2 plus a constant. Except in the 5-QTL scenario, realized accuracies with all WssGBLUP procedures were higher compared with those with BayesB and C. Noise in Manhattan plots was small with 5 and 100 QTLs but large with 500 QTLs. In the third and fourth studies, (co)variance components and prediction accuracy in linear and threshold, univariate, bivariate, and multivariate models were compared using ssGBLUP and BLUP methods for disease traits of binary or categorical nature. Uni- and multivariate threshold models surpassed linear models in obtaining higher heritabilities. A univariate threshold model surpassed a linear model in predicting (G)EBVs. Bivariate models and the ssGBLUP method did not have an advantage over univariate models and the BLUP method, respectively.