Adjusting Genomic Relationship Matrices in Single-Step Genomic BLUP for Crossbred Evaluations

Abstract Text: Different breed-specific genomic relationship matrices (GB) were compared to the standard across-breed genomic relationship matrix (G) used in single-step genomic evaluations. Datasets were simulated that resembled a terminal-cross population. Two purebred lines were separated by 50 generations. Three scenarios considered selection based on high EBV, high phenotypes, and no selection. The datasets used for evaluations contained phenotypes and pedigrees for the last 15 generations and genotypes for the last 8 generations of purebreds. Data on F1 animals were from a single generation. Number of purebred parents genotyped varied from 3100 to 3300 depending on the scenario, and number of genotyped F1 was 1200. The heritability for the simulated trait was 0.30. Testing involved four genomic matrices: GB1 considered specific allele frequencies (AF) for each pure and crossbred; GB2 used AF for crossbred calculated based on AF from the 2 purebreds; GB3 and GB4 had AF as in GB2 and GB1, respectively; however, each element was scaled by breed-specific scaling factors. Across-breed and breed-specific correction factors for G and all GB were also used to account for the non-random genotyping caused by selection. The validation was done in F1 animals and parameters of the regression of TBV on GEBV were used to assess the accuracy of evaluations. For G and all GB, coefficients of determination (R²) and regression were higher when no artificial selection was applied. When no correction factor was used, R² for G, GB1, GB2, GB3, and GB4 for EBV selection were 0.33, 0.03, 0.37, 0.37, and 0.03, respectively; for mass selection were 0.23, 0.33, 0.37, 0.38, and 0.33, respectively; for no selection were 0.47 for G and 0.46 for all GB.  However, after using breed-specific correction factors, the difference between G and GB was reduced and GB1 and GB4 gave similar results to G (R² = 0.40 under EBV and mass selection; R² = 0.47 under no selection), while GB2 and GB3 had slightly worse performance. Most unbiased predictions were with G and the correction factor applied, which regressions were close to 1.0 for the purebreds and from 0.65 to 1.0 for the crossbreds; the highest inflation was with the EBV selection and no phenotypes on crossbreds. Breed-specific genomic matrices provide little benefits for genomic evaluations in a terminal cross model. The best performance is with standard G corrected for an average selection across breed types. 

Keywords: ssGBLUP, breed-specific, allele frequency