Evaluation of the Effects of Genotyping Scenarios, Heritability, and Marker Density on Genomic Prediction Accuracy and Response to Selection in Sheep

Introduction
 Improving the production and reproductive performance of livestock is one of the best ways to increase the productivity of the livestock sector, and it can be achieved through breeding. Advances in selection methods, enabled by available technologies, will lead to the selection of superior animals. The goal of sheep breeding is to produce sheep with better performance and traits desired by breeders. Sheep breeders do not invest much in genetic improvement, also because only a small portion of the overall benefits of genetic improvement return to the breeder, so breeders seek genetic improvement at the lowest cost. As a result, they must understand that genomic selection is possible with sufficient accuracy and reasonable cost. To address this problem, many studies have focused on designing genotyping strategies to optimize individual selection. Considering the importance of sheep breeding and improvement, this study aims to investigate the prediction accuracy of traditional methods of best unbiased linear prediction (BLUP), genomic BLUP (GBLUP), and single-step GBLUP (ssGBLUP) in sheep using simulated data with different levels of heritability and marker density based on phenotypic performance or estimated breeding values (EBVs), to demonstrate the benefits of genomic selection and encourage sheep breeders to use genomic selection.
Method
 In this study, a population of sheep was simulated using QMSim software, based on the forward-in-time method. To evaluate the impact of different genotyping scenarios, five scenarios were simulated, including four genotyping scenarios (G1-G4) and a reference scenario (BLUP) in which no genotypes were considered. Scenarios G1-G4 included genotyping all animals from the past 15 generations, 10,000 ewes randomly selected from the past 15 generations, 10,000 candidates from both sexes randomly selected from the past 15 generations, and 10,000 rams randomly selected from the past 15 generations, respectively. In each genotypic scenario, to investigate the effect of phenotypic record on prediction accuracy based on three assumptions (phenotypic record for both males and females, phenotypic record for females, and phenotypic record for males), different phenotype determination scenarios were considered in which the percentage of phenotypic record was 20, 60, and 100 percent. After simulating the data with QMSim, different genotype and phenotype scenarios were analyzed using R. Afterward, the outputs from R software were used to estimate breeding value (EBV) and genomic breeding value (GEBV) using DMU software (Madsen and Jensen, 2013), considering the true genetic variance, using BLUP, GBLUP, and multi-breed ssGBLUP methods (which do not require pedigrees and genotype determination for all animals). Accuracy in each scenario was calculated as the correlation between the true breeding value (TBV) and the EBV or genomic EBV (GEBV) in the 15^th generation.
Results
In this study, the accuracy of GEBV prediction was investigated across different evaluation methods, heritability levels, and marker densities in simulated sheep populations. In the BLUP method, breeding values are estimated from phenotypes and pedigrees. The accuracy of prediction by the BLUP method based on heritabilities of 0.1, 0.3, and 0.5 was 0.27, 0.31, and0.33, respectively. The use of genomic information increased the accuracy in all scenarios compared to the BLUP method. The average accuracy of GBLUP, based on heritabilities of 0.1, 0.3, and 0.5, was 0.27, 0.30, and 0.32 for 10K marker density; 0.34, 0.32, and 0.34 for 50K marker density; and 0.36, 0.33, and 0.34 for 70K marker density, respectively. For the ssGBLUP method, the average accuracy at 10K marker density for heritabilities of 0.1, 0.3, and 0.5 was 0.30, 0.33, and 0.34; at 50K marker density, it was 0.36, 0.35, and 0.38; and at 70K marker density, it was 0.37, 0.36, and 0.40. The results showed higher prediction accuracy with the ssGBLUP method than with the GBLUP method.
This indicates an improvement in the accuracy of GEBVs obtained using ssGBLUP compared to GBLUP. Increasing marker density from 10K to 70K had the greatest effect on improving the prediction accuracy of a trait with a heritability of 0.1. Among the genotypic scenarios studied, the lowest and highest prediction accuracies were observed in scenarios G1 (genotyping without considering the restriction) and G4 (selective genotyping of rams), respectively. Among the scenarios studied, there was no significant difference in prediction accuracy between scenarios G2 (selective genotyping of ewes) and G3 (selective genotyping of rams and ewes). When genotypic information from ewes and rams was used simultaneously (G3), the prediction accuracy was slightly higher. In other words, determining the combined genotype of ewes and rams, regardless of the ewe selection strategy, results in moderate predictive accuracy. Across all phenotypic scenarios, a higher percentage of phenotypic records led to greater prediction accuracy. This trend was similar across all scenarios, regardless of whether phenotypic records were available for both sexes or only one. In the 10K marker density, hypothesis 1 (the existence of a phenotypic record for ewes and rams) was superior. In the 50K and 70K marker densities, hypothesis 2 (the existence of a phenotypic record in ewes) was superior. Among all the genotypic and phenotypic scenarios studied, the highest prediction accuracy was associated with the 70K marker density, with a heritability of 0.1.
Conclusions
A major challenge for sheep production is sustainably raising sheep and producing enough protein to feed the growing human population. Given the important role that small ruminants play in the livelihoods of livestock farmers, the introduction of genomic selection should be financially beneficial for breeders. As a result, using genomic selection at the lowest cost and with the highest prediction accuracy can encourage livestock farmers and sheep breeders to produce sheep with improved performance and desirable traits. By determining the genotype and phenotype of the selection candidates, in other words, by correctly ranking the selection candidates that will be chosen as the parents of the next generation, maximum genetic improvement can be expected. Also, by selecting the appropriate genotype, statistical method, and marker density, greater prediction accuracy can be achieved at lower cost.