The impact of information quantity and strength of relationship between training set and validation set on accuracy of genomic estimated breeding values

Abstract

Recent advances in genomic selection are a revolution in animal breeding. A genome consisting 10 chromosomes each with 100 cM in length with 100 equally spaced markers (1 cM) were simulated. After 50 generations of random mating in a finite population (Ne = 100) in order to create sufficient linkage disequilibrium, population was expanded to two different population sizes of 500 and 1000. This structure was conserved until generation 59. Only females of generations 51 to 58 had phenotypic
records and were included in the training set. The generation 59 was assumed as juveniles without any phenotypic records (validation set). Two measures of heritability (h2 = 0.1 and h2 = 0.5) were considered.
Each simulation was replicated 10 times and results were averaged across replications. The results showed that using individuals of more recent generations in training set led to higher accuracy of genomic estimated breeding values (GEBVs) than individuals from more distant generations. However, increase in the amount of phenotypic records in training set even from individuals of older generations will increase accuracy of GEBVs. Number of phenotypic records in training set was shown to have
important role in accuracy of GEBVs especially for low heritability traits.