Including Dominance Effects in Genomic Selection Regression Models with Different Priors - Pinus Taeda

In genomic prediction, the proper partition of the various sources of genetic variation (additive, dominance and epistasis) is desired for several applications, such as the use of models for mate-pair allocation. The objective of this study was to analyze the effect of including dominance effects in genomic selection regression models, using different priors. Model performance was assessed from 6-year total tree height measurements and rust resistance data obtained from a structured population of Pinus taeda composed of 923 individuals genotyped with 4722 SNPs. We also simulated a population derived from a pine breeding program, originated from selections made in a natural population with Ne=10,000, under a neutral model. Six traits (three dominance levels and two distributions of QTL effects) were simulated. For height, the inclusion of dominance effects increased the prediction ability, while for rust resistance the additive models provided a better fit. In the simulated traits, the accuracy was improved only for traits with high dominance. For polygenic traits, the priors had similar results, while for oligogenic traits BayesA and BayesB were superior. The results agree with those reported in the literature, and show that even with simulated traits with high dominance, the estimation of dominance effects is not as accurate as the estimation of additive effects. Further investigation of experimental precision and/or population size could be used to improve the accuracy of predicting dominance effects.