Effects of using superior animals in training set on the accuracy of genomic evaluation for high and low heritability traits in dairy cattle

Genetic improvement of dairy cattle has been successful during the second half of the 20th century, using pedigree and performance data. Genomic selection can improve the rate of this progress by reducing generation interval and increasing the accuracy of estimated breeding values at birth. There are many issues about using genomic evaluation. In the current study, some strategies were compared to inspect the effects of using superior animals as training set on the accuracy of genomic selection for high and low heritability traits.
For each individual, three chromosomes each with 1000 equally-spaced SNPs (each 0.1 cM one SNP) were simulated. For 50 generations, random mating was done to create enough linkage disequilibrium. In generation 51, the population size was expanded to 4000 individuals. There was this population size, until generation 55. In these generations, 5% of superior bulls and 80% of superior cows from each generation were used as the parents of the next generation. Animals in generation 55 (generation five) were validation group and animals in generations 51-54 (generations one to four) were training set. In strategy one (or first strategy), superior and inferior animals, in strategy two (or two strategy), superior animals and in strategy three (or two strategy), a random sample of the population were used as the training group. This research was done for traits with low (0.1) and high (0.5) heritabilities.
The results showed that the accuracy of genomic evaluation in strategy one was better than strategy two and strategy three. When the ratio and the number of animals in the training set were increased, the difference between the three strategies tended to decrease. For example, for traits with heritability of 10%, using 20% of animals from generations one to four, resulted in 29% difference between strategy one and two; in the same situation, using 40% of animals resulted in 14% difference between these strategies. The accuracy of genomic evaluation for the traits with high heritability of all strategies was higher and the difference between strategies was lower for these traits. For example, when 40% of animals of generations one to four were considered as training set, the difference between strategy 1 and 2, for trait with heritability of 0.1 was 14% and for trait with heritability of 0.5 was 4%.
Using only superior animals in the training group could decrease the accuracy of genomic evaluation. This effect was lower for high heritable traits. Using more animals would alleviate this effect by increasing total genetic variation explained by markers.