Selective genotyping strategies for genomic selection for host resistance against pancreatic disease in A. salmon

Genomic selection has become an increasingly important tool for improving aquaculture breeding programs, but the high cost of genotyping has been a barrier. Selective genotyping of phenotypically contrasting individuals is an approach that can reduce genotyping costs without significantly compromising the prediction accuracy. The aim of the study was to assess the effect of selective genotyping strategies on prediction accuracy and bias using linear and nonlinear genomic models in silico. The genotype and phenotype data from 7298 individuals belonging to 348 full-sib families were used from Mowi's Atlantic salmon commercial breeding population. A binary phenotype was generated from a challenge experiment to measure resistance to pancreatic disease (PD). Five selective genotyping strategies were tested: Full Genotyping (FG), Top-Bottom Genotyping (TBG), Minor Category Genotyping (MCG), Random Across family genotyping (RAFG), and Random Within family Genotyping (RWFG). Use of genomic linear models, GBLUP families, and non-linear models (Bayesian approaches) compared within the context of selective genotyping strategies. The prediction accuracies of the different genotyping strategies ranged between 0.44 and 0.87. Compared to the FG strategy, a 3–8% reduction in prediction accuracy was observed for the TBG, RAFG and RWFG approaches for the various genomic models. Differences among the TBG, RAFG and RWFG genotyping strategies were generally small. However, when the number of selected sibs for genotyping becomes below 8 per family, genotyping of the most phenotypically extreme individuals within a family is more advantageous. The Bayesian methods exhibited slightly higher prediction accuracy when 8 or more siblings per family were genotyped. However, as the number of training siblings per family decreased below 8, the single-step GBLUP methods performed slightly better. Adoption of single-step GBLUP has proven effective in mitigating the reduction in prediction accuracy and the increase in bias when selectively genotyping a subset of test siblings within the RAFG, RWFG, and TBG strategies.