Sensitivity of genomic single and multi-trait prediction accuracy to genetic architecture of the traits

Current research was carried out to study accuracy of single and multi-trait genomic prediction for two traits with 0.1 and 0.6 heritability; four 10, 50, 150 and 500 underlying QTL numbers for each trait and five normal, uniform, t, gamma and Laplace distributions for QTL effects using stochastic simulation. A genome with 10 Morgan lengths was simulated and 10000 markers were evenly spaced within it. In each scenario QTLs was randomly distributed on genome and their substitution effects was drawn from one of mentioned distributions. In order to create linkage disequilibrium between markers and QTLs genetic drift was used for 50 generations. In generation 51, called reference population, number of individuals increased to 1000 animals and phenotype of each of these animals was simulated. Marker effects were estimated in reference population using single and multi-trait best linear unbiased prediction (BLUP) methods. By using these estimated markers effects and genotype of markers for juvenile animals in generation 52, called validation population, genomic breeding values were predicted and different scenarios of this study were compared using accuracy of the predictions. Results of this study showed that in all scenarios high heritable trait had higher accuracy than low heritable trait. As well as, multi-trait BLUP increased nearly 10 percent accuracy of breeding values for low heritable trait, when genetic correlation between traits was 0.6. In all studied scenarios accuracies of genomic breeding values not affected by changes in number of QTLs and different distribution for QTLs effects. Therefore, multi-trait BLUP method without sensitivity to genetic architecture of traits can be used for genetic evaluation of low heritable traits.