17
Improving the Accuracy of Genomic Prediction of Milk Fat Yield in the New Zealand Holstein Friesian Population

Tuesday, March 18, 2014: 9:45 AM
312-313 (Community Choice Credit Union Convention Center)
Melanie K. Hayr , Iowa State University, Ames, IA
Mahdi Saatchi , Iowa State University, Ames, IA
Dave Johnson , LIC, Hamilton, New Zealand
Dorian J. Garrick , Iowa State University, Ames, IA
Abstract Text:

This study investigated the effect of including a QTL for milk traits, DGAT1, in calculating Direct Genomic Values (DGVs). Illumina SNP50 (50k) genotypes and Deregressed Estimated Breeding Values (DEBVs) for fat yield were provided by LIC for 5,661 Holstein Friesian cows and 2,287 bulls. DGAT1 genotypes were provided for 1,133 cows and 655 bulls, with DGAT1 genotype imputed for the remaining cattle using BEAGLE. Four models were run in GenSel using Bayes B method and 5-fold cross-validation with 2.5% of SNPs assumed to have an effect on the trait: 1) a model relying on linked 50k markers to pick up the effect of DGAT1; 2) a model with 50k markers and DGAT1 dosage fit as a random covariate; 3) a model with 50k markers and DGAT1 genotype fit as a fixed class; and 4) a model with 50k markers and DGAT1 dosage fit as a fixed covariate. These models were run separately for males and females and each sex was run twice, once with only animals with DGAT1 directly genotyped and then with all animals. Accuracy was defined as the correlation between DEBV and DGV while bias was represented in terms of the regression coefficient of DEBV on DGV. Performance was very similar in models 1 and 2 while results for models 3 and 4 were also very similar. Models 3 and 4 performed better than models 1 and 2. When all animals were included, the models with 50k markers plus DGAT1 as a fixed class or a fixed covariate performed equivalently. When only animals directly genotyped for DGAT1 were analyzed the model with 50k markers plus DGAT1 as a fixed covariate had the lowest bias, while the model with 50k markers plus DGAT1 as a fixed class had the highest accuracy. These results were consistent across both sexes. These results suggest that including DGAT1 genotype as a fixed class or a fixed covariate when calculating DGVs both increases accuracy and reduces bias.

Table 1. Regression coefficients and correlations between DEBV and DGV

Sex

Model

Direct DGAT1

Direct and Imputed DGAT1

b

r

b

r

Males

50k

1.246

0.553

1.012

0.697

50k+DGAT1(Random Covariate)

1.246

0.552

1.010

0.696

50k+DGAT1(Fixed Class)

1.191

0.660

1.007

0.737

50k+DGAT1(Fixed Covariate)

1.055

0.536

1.008

0.737

Females

50k

1.141

0.399

1.048

0.385

50k+DGAT1(Random Covariate)

1.140

0.400

1.045

0.384

50k+DGAT1(Fixed Class)

1.083

0.503

1.039

0.453

50k+DGAT1(Fixed Covariate)

1.038

0.463

1.040

0.455

Keywords: Dairy, DGAT1, Genomic Prediction