Impact of preferential treatment of elite cows on accuracy of genomic predictions in a simulated dairy cattle population

Abstract

<p>Implementation of genomic selection using a cow training population is of interest for countries with no or limited progeny-tested bull data. However, preferential treatment of elite cows may cause bias in genomic predictions. The objective of this study was to investigate the impact of using preferentially treated cows in the training population on accuracy and bias of genomic predictions. A population undergoing a four-pathway selection strategy similar to that in dairy cattle was simulated. Two traits with low (0.05) and moderate (0.3) <i>heritability were considered</i>. The training population consisted of between 2,500 and 20,000 randomly selected cows. Preferential treatment (PT) was simulated and introduced to 5, 10 and 20% of elite cows. Preferential treatment of elite cows resulted in lower accuracy of predictions compared to the control scenario without PT. The accuracy of genomic predictions in the control scenario ranged from 0.72 to 0.83 and from 0.75 to 0.86 for traits with heritability of 0.05 and 0.3, respectively. When the training population included 20% preferentially treated cows, corresponding accuracies decreased to 0.68 to 0.80 and 0.64 to 0.77, respectively. Training on cows with PT resulted in upward bias of genomic predictions (regression coefficient 0.79 and 0.45 for traits with heritability of 0.05 and 0.3, respectively). Generally, using cow data in the training population is an attractive way to implement genomic selection for countries with no or limited progeny-tested bull data. However, further investigation is needed to adjust for or remove bias due to potentially presence of preferential treatment.</p>