Abstract:

Accurate genomic predictions underpin the success of genomic selection-based breeding programs. Therefore, efforts to compute and improve prediction accuracy (ACC) are imperative. Here we use theoretical expectations and simulation to highlight the linear relationship that exists between ACC and phenotypic differences in pre-defined percentiles. We found that an ACC of 0.4 yielded a phenotype difference of one standard deviation (SD) units between the top and bottom quartiles. In addition, a seemingly moderate increase in ACC of 0.05 (absolute value) translates into an increase of 0.127 SD units between top and bottom quartiles and to a 2.54 SD units’ difference when extrapolated to a perfect ACC. These theoretical expectations were validated with results from a dataset of 3,408 Australian Angus steers with genomic predictions for a range of feedlot and carcase traits. We conclude that moderate predictions accuracies of ~0.4 are sufficient to effectively draft animals into quartiles.