Comparison of Single and Multi-Step Bayesian Methods for Predicting Genomic Breeding Values in Genotyped and Non-Genotyped Animals- A Simulation Study

The purpose of this study was to compare the accuracy of genomic evaluation for Bayes A, Bayes B, Bayes C and Bayes L multi-step methods and SSBR-C and SSBR-A single-step methods in the different values of π for predicting genomic breeding values of the genotyped and non-genotyped animals. A genome with 40000 SNPs on the 20 chromosom was simulated with the same distance (100cM). The π values that maximized the prediction accuracies in BayesC were 0.980 and 0.995 for the normal and gamma distributions of QTL, respectively, and were also used in SSBR-C method. Genomic prediction accuracy in the SSBR-C (π = 0.99) method was higher than multi step methods from 0.02 to 0.09 for gamma distribution. Results showed that considering mixture distribution and use of phenotype, genotype and pedigree information simultaneously, the SSBR-C (π = 0.99) method had higher accuracy than other methods and is considered a better choice in this scenario. Moreover, both single and multi-step methods showed similar prediction accuracy when the genetic architecture appeared to approach the normal distribution. Furthermore, SSBR-C (π = 0) method appeared to be more reliable choice that was due to regressions of true breeding value on estimated breeding value close to one in normal distribution. Generally, GEBV accuracy decreased as the distance increased between validations and training set, which was more sensitive for non-genotyped individuals compared to genotyped individuals.