Bacterial cold water disease (BCWD) causes significant mortality and economic losses in salmonid aquaculture. Previously, we reported high genomic prediction accuracy of 0.72 for BCWD resistance in rainbow trout using the 57 K Axiom SNP array. Due to the high cost of phenotyping and genotyping in rainbow trout breeding programs, it is paramount to know if acceptable genomic prediction accuracy can be obtained in the subsequent generation without retraining the prediction model. In the current study, we found that the accuracy of genomic prediction without model retraining in the subsequent generation was reduced from 0.65 and 0.61 to 0.56 and 0.53 using a lower density array of 10 K SNPs and a panel of 49 QTL-linked SNPs, respectively. Although markedly lower than the genomic prediction with model retraining, the prediction accuracy without retraining was better than the pedigree-based model (PBLUP) with retraining (0.48) and substantially higher than the PBLUP model accuracy without retraining (0.22). We conclude that genomic selection provides better prediction accuracy than the traditional PBLUP model even when ‘skipping’ one generation of model retraining. The weighted single-step GBLUP (wssGBLUP) and single-step Bayesian multiple regression BayesB (ssBMR-BayesB) had higher genomic prediction accuracy than single-step GBLUP (ssGBLUP), which is consistent with the oligogenic inheritance of BCWD resistance in this population. Imputation from the 10 K array genotypes back to the 57 K array genotypes did not improve the accuracy of genomic prediction, likely due to the high linkage disequilibrium in rainbow trout aquaculture breeding populations and the oligogenic architecture of BCWD resistance.

细菌性冷水病 (BCWD) 在鲑鱼养殖中造成显着的死亡率和经济损失。之前，我们使用 57 K Axiom SNP 阵列报告了虹鳟 BCWD 抗性的高基因组预测准确度为 0.72。由于虹鳟育种计划中表型和基因分型的成本很高，因此在不重新训练预测模型的情况下，了解是否可以在后代中获得可接受的基因组预测准确性至关重要。在当前的研究中，我们发现，使用低密度的 10 K SNP 阵列和一组 49 个 QTL 连接的 SNP，在后代中无需模型再训练的基因组预测准确性分别从 0.65 和 0.61 降低到 0.56 和 0.53 . 虽然明显低于模型再训练的基因组预测，没有再训练的预测精度优于有再训练的基于谱系的模型 (PBLUP) (0.48)，并且大大高于没有再训练的 PBLUP 模型精度 (0.22)。我们得出结论，即使在“跳过”一代模型再训练时，基因组选择也能提供比传统 PBLUP 模型更好的预测准确性。加权单步GBLUP（wssGBLUP）和单步贝叶斯多元回归BayesB（ssBMR-BayesB）比单步GBLUP（ssGBLUP）具有更高的基因组预测准确性，这与该人群BCWD抗性的寡基因遗传一致。从 10 K 阵列基因型推算回 57 K 阵列基因型并没有提高基因组预测的准确性，