212 Genomic prediction of carcass average daily gain, fat and loin depth in three-way crossbred pigs including information collected on purebreds
Abstract The purpose of this study was to predict three-way crossbred performance for carcass traits using different crossbred/purebred reference populations. Carcass measures (average daily gain, back-fat and loin depths) were collected in 4,893 three-way-cross individuals (CB individuals, 1,252 being genotyped). Live measures of body weight and tissue deposition were collected on 3,050 purebred Duroc individuals (PB individuals, 941 being genotyped), paternal-half-sibs (PHS) of the CB individuals. Models’ predictive performance was tested via 4-fold cross-validation. The basic model included CB phenotypes from the training set without inclusion of genomic information (i.e. pedigree BLUP). We also sequentially included: 1) CB genotypes; 2) PB phenotypes and genotypes for the training families (PBt); 3) PB phenotypes and genotypes for the validation families (PBv). Variance components (heritabilities and genetic correlations between CB and PB traits) were not estimated but fixed at different values within a plausible interval, the combination of such parameters that gave the best predictive ability was considered for that model. Results reported pedigree prediction of CB traits to show about 0.25 accuracy (correlation between breeding value and adjusted phenotype) for the three traits. The inclusion of CB genotypes was beneficial, with an increase ranging from 25 to 50% (depending on the trait) compared to pedigree prediction. When PBt genotypes and phenotypes were included, prediction accuracy dropped to almost null accuracy. When PBv genotypes and phenotypes were included, predictive performance was better than models that included CB information only. Results suggest that PB information can improve selection accuracy for CB traits, with the condition PB are PHS of the CB in validation. Otherwise, inclusion of PB information from the training set can be detrimental. CB genotypes, on the other hand, always improve prediction accuracy. We can conclude that reference populations aimed at improving CB performance should include phenotypes and genotypes from these individuals.