20
Genomic prediction accuracies of residual feed intake (RFI) and component traits of feed efficiency in pigs divergently selected for RFI
The objective of this work was to evaluate the accuracy of genomic predictions for residual feed intake (RFI) and component traits of feed efficiency in two lines of pigs selected for high (H-RFI) and low (L-RFI) RFI (defined as observed feed intake minus expected feed intake). Phenotypic and genotypic data on 994 L-RFI pigs from generations 0 to 10 of selection and 698 H-RFI pigs from generations 4 to 10 were used in this study. Pigs were on-test at 95±15 days of age (39.8±10.9 kg BW) and off-test at 202±17 days of age (115.8±7.4 kg BW). Phenotypic data included average daily feed intake (ADFI), average daily gain (ADG), backfat depth (BF), loin muscle area (LMA), feed conversion ratio (feed-to-gain ratio; FCR), and RFI (ADFI adjusted for ADG, LMA, BF, and metabolic BW). All animals were genotyped using the Illumina PorcineSNP60 BeadChip, and after quality control, 51,098 single nucleotide polymorphisms (SNP) were used for analyses. For genomic prediction, one line was used as the training dataset and the other for validation. Marker effects were estimated for all SNPs using the genomic prediction methods Bayes-B and Bayes-C, fitting as many SNPs as degrees-of-freedom available after accounting for fixed-effects in the model. Accuracy of genomic prediction was calculated in the validation set as the correlation of genomic predictions with phenotype pre-adjusted for fixed effects, divided by the marker-based heritability. Results are in Table 1. Overall, accuracies were sizeable, considering that training and validation were separated by at least 5 generations of selection, and both Bayesian genomic prediction methods resulted in similar results. The only trait with inconsistent results was FCR, in which Bayes-B resulted in low accuracy (0.06) and Bayes-C in moderate accuracy (0.28) when training on L-RFI and validating on H-RFI. These results suggest that RFI and component traits can be predicted between divergent lines selected for RFI using high-density SNP genotypes. Financial support from AFRI-NIFA grant #2011-68004-30336 is appreciated.
Table 1. Genomic prediction accuracies when training in one RFI line and validating on the other
Method |
Training |
ADFI |
ADG |
BF |
LMA |
FCR |
RFI |
Bayes-B |
L-RFI |
0.36 |
0.36 |
0.24 |
0.36 |
0.06 |
0.31 |
|
H-RFI |
0.39 |
0.32 |
0.29 |
0.22 |
0.18 |
0.34 |
Bayes-C |
L-RFI |
0.35 |
0.34 |
0.23 |
0.30 |
0.28 |
0.32 |
|
H-RFI |
0.40 |
0.32 |
0.20 |
0.26 |
0.14 |
0.43 |
Keywords: Feed Intake, Genetic Improvement, SNP