Evaluation of antedependence model performance and genomic prediction for growth in Danish pigs

Abstract Text: The widely used genomic prediction models such as GBLUP, BayesA/B models all assume marker effects independent. Bayesian antedependence models extend this by estimating correlated marker effects, arising from linkage disequilibrium between markers. In this study we compare model fit and complexity, as well as prediction accuracy between antedependence models and other models applied to Danish Duroc pig data, including 29,567 SNPs. The results showed that anteGBLUP model and other conventional models did equally well in prediction. DIC for the models showed that anteBayesA and double-anteBayesA models had better fit, but higher number of effective parameters, and lower accuracy. In conclusion, the simple antedependence model works well for genomic prediction, but more complex antedependence models may be interesting to estimate marker effects correcting for LD structure. The DIC appears a good indicator of prediction accuracy.  

Keywords:

antedependence model                                     

genomic prediction

model complexity and fit