Study On The Methods Of Genomic Selection For Meat Sheep By Simulation

With the availability of high and low density SNP chips for research and production practice, and advances in computer technology, genomic selection (GS) is becoming a hot topic in the research of animal breeding and genetics.In this study, two different populations were simulated with daily gain (h2=0.21) and lean meat yield (h2=0.34) of meat sheep. According to the genome of sheep published by the NCBI, we simulated40000single nucleotide polymorphisms (SNPs), then selected500,1000,3000,500010000SNPs randomly and all SNPs as the density of SNP chips. Genomic estimated breeding value (GEBV) was respectively predicted by three different methods GBLUP, BayesCPi and Bayesian LASSO in different SNP chip densities and reference population sizes, and the accuracy of GEBVs and persistence of GS were compared and analyzed. The results are as follows:1. The accuracy of additive genetic relationship estimated by wide genome markers is higher than that of classical numerator relationship matrix. Marker densities ranged from500to40000were all suitable for construction of the whole genomic relationship matrix (GRM).2. The maximum effect of markers and the number of markers with large effects estimated by BayesCPi were higher than those estimated by Bayesian LASSO when the marker density was40000. Nearly the same location of marker loci with the maximum effect was obtained by BayesCPi and Bayesian LASSO. Increasing the size of reference population can increase the estimates of SNP effects and the number of SNPs with large effects.3. The accuracy of GEBV estimated by BayesCPi was slightly higher than that estimated by GBLUP and Bayesian LASSO when the density was40000,10000,5000and3000. However, slightly lower accuracy of GEBV was obtained by BayesCPi with the low marker density of1000and500. The BayesCPi method was much highly influenced by marker density than the other two methods as suggested by regression coefficients of true breeding value on estimated breeding value. The Bayesian LASSO was robust as indicated by high accuracies and little changes of GEBVs estimated with different marker densities. It was not robust for GBLUP method under low marker densities.4. The accuracy of GEBV was increased with the size of reference population. If the SNP effect estimated in reference population was used to estimate the GEBV of reference population itself, GEBV accuracy increased with the genetic variation within the reference population. The influence of genetic variation on GEBV accuracy was greater than that of population size. When SNP effect estimated in reference population was used to estimate the GEBV of candidate population, accuracy of GEBV and persistence of GS were improved with increased population sizes.5. The accuracy of GEBV for daily gain with moderate heritability was higher than that for lean meat yield with high heritability when the reference population was smaller (348or355). With the increased population sizes (700～1700), the accuracy of GEBV for daily gain was lower than that for lean meat yield.These results obtained in this thesis provide scientific bases for the practical genetic improvement of meat sheep by using genomic selection.