The accuracy of genomic prediction for viral nervous necrosis and vibrosis disease resistance in atlantic cod

The objective of this study was to estimate accuracy of genomic prediction for disease resistance to viral nervous necrosis and vibriosis using sparse and genome sequence SNP-data in Atlantic cod. The disease challenge test data of viral nervous necrosis and vibriosis used in this study were obtained from the National Atlantic cod breeding program which is running in Tromsø, Norway and we used disease challenge test data of year-class 2009 for both traits. Disease resistance for both traits was measured as survival at a fixed point in time and assessed as a binary variable. We obtained the result of challenge test data of 707 and 728 individuals for viral nervous necrosis and vibriosis respectively. The individuals came from75 full-sib and half-sib families for both diseases and the number of individuals per family varied from 7 to 20 (average of 9.7) in viral nervous necrosis, and 6 to 10 in vibriosis. On top of pedigree information of 1,743 individuals, three genotype data sets were used in this study, and based on these data sets three different genomic relation matrices were calculated. These were SPARSE8 (genotype data of 283 SNP markers at chromosome 8 of 1,743 individuals), SPARSE GENOME (1,577 individuals’ genotype data of 8,658 SNP markers across the entire genome) and DENSE8 (imputed high density genotypes (759,270 SNPs) of chromosome 8 of 1,743 individuals). The genomic relation matrices were used in the GBLUP with polygenic models to estimate the variance components which were explained by the genomic information, and the genomic estimated breeding values using ASReml software. Fivefold within-family cross validations were carried out by randomly masking 20% of phenotypic records within each family in order to evaluate the accuracy of prediction for the viral nervous necrosis disease trait. Each observation was masked once and 141 phenotypes were masked in the first, second and third cross validation tests, whereas 142 phenotypes were masked in the fourth and fifth cross validation tests. Finally, the phenotypic values of the masked individuals were predicted based on the 566 or 565 phenotypic observations of the unmasked individuals. In the case of a between- families cross validation test, the phenotypic values of 20% of the families were masked at a time and their phenotypic values were predicted from the other families’ phenotypic values. A total of 15 families were masked in each cross validation and the total masked phenotypes were 140, 1