Effect of the Reference Population Composition on Genomic Predictions of Small Dairy Populations

The inclusion of genomic information in animal improvement programs, brought the possibility of calculating direct estimated breeding values (DGBV). The estimation of DGBV requires validation because the formulae are calculated based on a reference population. The objective of this study was to measure the effect of the reference population composition on DGBV and its reliabilities in small populations. In the analysis, 2,011 genotyped animals (previously imputed to 45,109 markers) were included, as was pedigree information of genotyped animals and their contemporaries (42,154 individuals) and available phenotypes of all females. Phenotypic information consisted of milk yield in Kg adjusted to 305 days. To estimate marker effects, an animal model was used including herd-year-season, sire-herd interaction, animal and residual effect as factors. The BLUPF90 software was used, which implements a single step procedure. Three reference population scenarios were evaluated A) Cross validation, creating 10 randomly groups (each one included 10% of the genotyped animals), B) Selection of 25 % upper and lower animals under the normal distribution based on phenotype information and C) 50% of the genotyped animals selected randomly. Averages and standard deviations of DGBV were 130±506, 52±512 and 93±501 and reliabilities were 0.66, 0.58 and 0.58 for scenario A, B and C respectively. Correlations of DGBV between scenarios A & B and A & C were 0.99. It is concluded that in small populations, the composition of the reference population for validation has little effect on DGBV estimated but it can decrease the average reliability.