Comparative analysis of the GBLUP, emBayesB, and GWAS algorithms to predict genetic values in large yellow croaker (Larimichthys crocea).

backgroundThe advances of sequencing technology accelerate the development of theory of molecular quantitative genetics such as QTL mapping, genome-wide association study and genomic selection. This paper was designed to study genomic selection in large yellow croaker breeding. The aims of this study were: (i) to estimate heritability values of traits in large yellow croaker; (ii) to assess feasibility of genomic selection in the traits of growth rate and meat quality; (iii) to compare predictive accuracies affected by different algorithms and training sizes, and to find what training sizes could reach ideal accuracies; (iv) to compare results of GWAS with genomic prediction, and to assess feasibility of pre-selection of significant SNPs in genomic selection. 500 individuals were tested in the trait of body weight and body length, while 176 were tested in the percentage of n-3 highly unsaturated fatty acids (n-3HUFA) in muscle. GBLUP and emBayesB were used to perform genomic prediction.

resultsGenotyping-By-Sequencing method was used to construct the libraries for the NGS sequencing and find ~30,000 SNPs. Heritability estimates were 0.604, 0.586 and 0.438 for trait of body weight, body length and n-3HUFA, respectively. The predictive abilities estimated by GBLUP showed higher than that by emBayesB in traits of body weight and body length. However, the result was just the opposite in n-3HUFA. According to fit the curve of predictive accuracy, we estimated that at least 1000 individuals in training set could reach an accuracy of 0.8 in body weight and body length. GBLUP, emBayesB and GWAS could not always find significant SNPs associated with phenotypes consistently. Significant SNPs were selected by emBayesB could obtain the largest proportions to explain total additive genetic variances.

conclusionsThis research showed that genomic selection was feasible in large yellow croaker breeding. We suggest doing a test before deciding to use which algorithm in specific trait in genomic prediction. We estimated required training sizes to reach ideal predictive accuracies and assessed feasibility of pre-selection of SNPs successfully. Because of high mortality rate of fish and high cost in genomic sequencing, genomic selection may be more suitable for applying on some traits which cannot be measured on candidates directly.