Improving genomic prediction accuracy for harvest traits in Asian seabass (Lates calcarifer, Bloch 1790) via marker selection

Genomic selection provides an efficient strategy to improve growth performance and processing traits in aquaculture. However, the high cost of genotyping animals using high-density SNP arrays may prevent its application in small-sized breeding programs. The aim of this study was to evaluate the utility of low-density SNPs for genomic prediction of harvest traits in Asian seabass. A total of 1047 fish were genotyped for 29,521 single nucleotide polymorphisms. Genome-wide association studies (GWAS) using the weighted genomic best linear unbiased prediction (wGBLUP) procedure revealed polygenic nature of harvest traits. GBLUP and BayesB were implemented using five sets of informative SNPs specific to each trait (ranging from 500 to 12,000) and five sets of evenly-spaced markers at the same density. Prediction accuracies were evaluated using five-fold crossvalidation with five replications. Using trait-specific panels with the largest effects of 500 to 2000 SNPs under GBLUP, prediction accuracies increased substantially for all traits relative to those from full SNPs, i.e., by -70% for body weight (BW), by -100% for standard length (SL), by -97% for fillet weight (FW) and by -660% for fillet yield (FY). A similar trend in prediction accuracy was observed from BayesB. Improvement in accuracy was marginal for evenly-spaced markers compared to full SNPs, i.e., by -25% for BW, SL, FW and by -30% for FY. Close relationship between training and validation fish appeared to be a major contributing factor to the favorable performance of prediction models for the within-family selection scheme of Asian seabass.