Examining Different Models of Gene Action in Genomic Evaluation

The purpose of this study was to compare different models of gene action in genomic selection using two statistical methods including Genomic Best Linear Unbiased Prediction (GBLUP) and Bayesian Least Ab-solute Shrinkage and Selection Operator (BLASSO). Therefore, three gene action models including purely additive effects (A), additive effects plus dominant deviations (AD) and additive effects plus dominant de-viations plus epistasis interactions (ADE) were fitted to the data by each method (i.e., GBLUP or BLASSO, GBLUP-D or BLASSO-D and GBLUP-DE or BLASSO-DE, respectively). Real genotypic data of mice were used and, phenotypes were simulated from these real genotypes based on different number of quantita-tive trait loci (QTLs) and levels of broad-sense heritability (T1:20 QTLs and H2=0.4, T2:20 QTLs and H2=0.8, T3: 100 QTLs and H2=0.4, T4:100 QTLs and H2=0.8, T5:200 QTLs and H2=0.4, T6:200 QTLs and H2=0.8). BLASSO recorded a higher prediction accuracy (PA) than GBLUP. This increase in PA was greater when the number of QTLs was low (20 QTLs), and this advantage decreased with the increase in the number of QTLs. When traits were under the control of a small number of QTLs, the PA of model A was slightly higher than AD and ADE models. The GBLUP-D or BLASSO-D methods showed a lower bias compared to GBLUP or BLASSO methods, and this bias was minimized in GBLUP-DE or BLASSO-DE methods for all traits.