Accuracy of within-family multi-trait genomic selection models in a sib-based aquaculture breeding scheme

Genomic selection has a great potential to increase genetic gain in aquaculture breeding; however, its implementation is hindered by a high genotyping cost due to a large number of individuals to genotype. Withinfamily genomic selection, which could utilize low-density markers and pedigree information, is suggested as a cost-effective way of implementing genomic selection in aquaculture. In this study, a single trait genomic model (STGM) is compared with a multi-trait genomic model (MTGM) for prediction of within-family genomic breeding values in a simulated sib-evaluated aquaculture breeding scheme. Two traits, one with lower heritability (h(1)(2 )= 0.05) and another with higher heritability (h(2)(2) = 0.5) were simulated. Three genetic correlations (r(g) = 0.2, r(g) = 0.5 and r(g) = 0.8) and zero residual correlation were assumed between these two traits. Given these assumptions, genomic and phenotypic data were simulated for 100 full-sib families of size 100. From each family, 10 individuals were randomly selected as selection candidates and the number of tested sibs varied from 10 to 90 per family. Two scenarios were investigated: in scenario I, all reference sibs were phenotyped for both traits, whereas in scenario II half of the reference sibs measured for trait I and the remaining half were measured for trait II. These scenarios were also compared under four SNP densities (10, 20, 50 and 100 SNP/Chr). For both STGM and MTGM, prediction accuracies increased as the number of tested sibs per family increased from 10 to 90, however, the rate of increase was higher for STGM. Compared to STGM, use of MTGM increased the accuracy by up to 71% in scenario II and by up to 58% in scenario I for the low heritability trait when the genetic correlation between the traits was 0.8. The highest improvement in accuracy was observed in scenario II when only 10 sibs were genotyped per family with 10 SNP/Chr. As the magnitude of the genetic correlation between the traits decreased, the relative gain in accuracy by implementing MTGM was reduced. The relative importance of MTGM also declined with the increase of the number of tested sibs per family and a similar trend, but with lesser magnitude, was observed with the increase of marker density. The results indicate that MTGM performs better than STGM for low heritability traits that are genetically correlated with high heritability traits. The advantage of multi-trait model was greater when both traits are not measured on the same group of individuals.