235 The impact of genotyping strategies and pedigree errors on single-step genomic prediction in sheep populations: a simulation study.

Abstract Genomic predictions provide more accurate estimated breeding values (EBV) in younger animals, enabling greater rates of genetic gain, but are still limited in sheep mostly because of high genotyping costs relative to individual animal values. Well-designed reference populations may, therefore, be particularly important in preventing biased genomic predictions. This study evaluated genotyping strategies by varying the proportion of genotyped animals considering different scenarios of pedigree errors (misidentified sires or missing information). Impact on EBV accuracy, bias, and dispersion was assessed using true breeding values and compared with Linear Regression (LR) statistics, mimicking real-world scenarios. We simulated a composite sheep population using the AlphaSimR package to mimic the formation of the Katahdin breed, and flock connectedness amongst flocks. The simulation included 16 flocks with an effective population size of 103, two traits with heritabilities of 0.35 and 0.10, and a 50K array. Breeding values were predicted with pedigree Best Linear Unbiased Prediction (BLUP) and Single-step Genomic BLUP (ssGBLUP). Eight flocks used BLUP EBV to select replacement animals while the other eight flocks used phenotypes. Scenarios included combinations of 0, 5, 10, 15, 20, 40, 80, or 100% males or females genotyped; 0, 5, 10, or 20% pedigree errors; and three genotyping strategies (random, highest BLUP EBV, or highest phenotypic values). The final population (18,717 animals) was divided into training (first three generations) and validation (last generation) sets to calculate EBV accuracies, biases, and dispersions. Furthermore, partial and whole data sets were created to compute the corresponding LR statistics. Correlations between true and LR statistics are shown in Table 1. The accuracy of genomic predictions improved significantly with random genotyping, outperforming phenotype and BLUP EBV-based strategies by up to 19%. Pedigree errors reduced accuracy, and increased bias and dispersion. Although both types of pedigree errors should be minimized, missing pedigree information impacted results more substantially than misidentified sires. Increasing the proportion of animals genotyped improved prediction metrics, with random genotyping yielding higher accuracies, lower biases, and dispersion closer to 1. Genotyping up to 10% of males before starting to genotype females enhanced the accuracy of the EBV. Compared to the true metrics, the LR metrics were suitable for validating genomic predictions, achieving moderate to high correlations. Dispersion decreased toward 1 (desirable) as accuracy increased, but when accuracy declined, dispersion increased beyond 1. When bias increased, dispersion increased. Genomic information mitigated some effects of pedigree errors. The ssGBLUP method was robust to pedigree errors and outperformed BLUP with a complete pedigree. Selective genotyping increased EBV bias and dispersion, and reduced prediction accuracy and long-term genetic progress. Similar conclusions were obtained for both trait heritability levels. These findings highlight the importance of well-designed genotyping strategies when implementing genomic selection in sheep.