The Impact of Genotyping Strategies on Accuracy of Genomic Prediction in Sheep Populations: Preliminary Results of a Simulation Study

Abstract Genomic predictions provide more accurate estimated breeding values (EBV) in younger animals, enabling greater rates of genetic gains over time. However, genomic selection in U.S. sheep is still incipient due to the higher genotyping cost relative to animal cost. Our objective was to compare genotyping strategies differing in the proportion of animals genotyped and the extent of pedigree error on prediction accuracy. These genetic evaluations were based on single-step Genomic Best Linear Unbiased Prediction. We used the AlphaSimR package to simulate a composite sheep population mimicking the Katahdin breeds’ formation. We then simulated four further generations based on the structure of Katahdin flocks engaged in the National Sheep Improvement Program. Thirty-five flocks were simulated with an effective population size of 170. Flock size ranged from 31 to 680 animals and, by design, was skewed towards smaller flocks. Average ram to ewe ratio, mortality rate per generation, and number of lambs born per ewe lambing was 1:16, 10%, and 1.8 lambs, respectively. We simulated a 50k SNP panel for a moderately heritable (0.30) trait. Flocks had low genetic connectedness. Selection criteria differed among flocks: 3 selected randomly, 10 selected based on highest EBV, and 22 selected based on highest phenotypes. All combinations of 5, 40, or 100% of males or females genotyped, 5 or 20% of animals missing pedigree, and 5 or 20% of sires misidentified were compared. We also considered a scenario with no pedigree errors (i.e., no missing pedigree or misidentified sires). Animals with the best phenotype were chosen for genotyping. The first three generations were used as the reference population (7,317 animals) to validate the 4th generation (2,055 animals). Each scenario was replicated 5 times. A linear model was used to estimate differences in prediction accuracy between scenarios, tested with Tukey’s test (alpha 0.05). For the scenario without pedigree errors, prediction accuracy ranged from 0.50 ± 0.06 for 5% of males and females genotyped to 0.75 ± 0.03 for 100% of animals genotyped (Figure 1). For scenarios with 5% pedigree errors, prediction accuracy ranged from 0.33 ± 0.02 for 5% of males and females genotyped to 0.71 ± 0.02 for 100% of animals genotyped. For the scenario with 20% pedigree errors, prediction accuracy ranged from 0.26 ± 0.06 for 5% of males and females genotyped to 0.67 ± 0.03 for 100% of animals genotyped. Across pedigree error scenarios, trends were similar. Importantly, genotyping 5% of one sex and 100% of the other resulted in similar accuracies to genotyping 100% of animals. In scenarios with few animals genotyped, accuracies were higher when errors in pedigree were avoided. These preliminary results reinforce the importance of genomics in correcting pedigree errors, with high levels of prediction accuracy when at least most animals of one sex are genotyped.