Consequences of ignoring dominance genetic effects from genomic selection model for discrete threshold traits.

The aim was to study the consequences of ignoring dominance effects from the genomic evaluation model on the accuracy, mean square error, bias, and dispersion of genomic estimated breeding values (GEBVs) for a discrete threshold trait. Also, the predictive performance of the parametric and non-parametric genomic selection models was compared. A genome consisting of 10 chromosomes, on which 10,000 bi-allelic single nucleotide polymorphisms (SNP) were distributed was simulated. In different scenarios, 100, 500, and 1000 SNPs were assigned to quantitative trait loci (QTL). For QTL effects, different distributions (normal, uniform, and gamma) were considered. While all QTLs were assigned additive genetic effects, in different scenarios, dominance effects were given to 0.00, 10, 25, 50, and 100% of QTLs. The GEBVs were estimated using the GBLUP, Bayesian methods, ridge regression variants, and machine learning methods. The criteria of the LR method such as prediction accuracy, mean square error, bias, and dispersion were used to analyze GEBVs. The results showed that if dominance effects were present, but not included in the model, caused to 14 to 31% decrease in the accuracy of the GEBVs. Also, the mean square error of GEBVs increased between 19 and 47%, following an increase in the percentage of QTLs with a dominance effect from 0.00 to 100%. In addition, the bias of GEBVs increased between 20 and 42%, and a 50% increase was observed in the dispersion of GEBVs by ignoring dominance effects. Results showed that when the genetic architecture of the traits is purely additive, or when dominance effects are present but ignored from the genomic evaluation process, GBLUP, BayesB, and ridge regression BLUP-method 6 (rrBLUPm6) had better predictive performance, and therefore recommended for genomic evaluation of discrete threshold traits. Machine learning methods, in particular regression tree, had poor predictive performance and were not recommended for genomic selection. In general, the results showed that not accounting for dominance effects when they are really present, leads to inaccurate, biased, and dispersed estimates of GEBVs, which will ultimately reduce the efficiency of genomic selection.