Weighted ssGBLUP Improves Genomic Selection Accuracy for Bacterial Cold Water Disease Resistance in a Rainbow Trout Population

The objective of this study was to compare methods for genomic evaluation in a Rainbow Trout (Oncorhynchus mykiss) population for survival when challenged by Flavobacterium psychrophilum, the causative agent of bacterial cold water disease(BCWD). The used methods were: 1)regular ssGBLUP that assumes all SNPs have the same variance; 2)weighted ssGBLUP(wssGBLUP) that gives more weight to SNPs that explain considerable portion of the genetic variance; both GEBV and SNP effects are updated iteratively; 3)BayesB that performs variable selection. The benchmark method was traditional BLUP. While BayesB used only phenotypes of genotyped animals, ssGBLUP considered all phenotypes, genotypes, and pedigrees jointly. Phenotypes for survival days and pedigrees were available for 4,004 and 5,104 individuals, respectively; whereas 2,490 animals were genotyped for 41k SNPs. Accuracy was the correlation between adjusted phenotype and GEBV divided by the square root of heritability, in a 5-fold cross validation. Accuracies obtained from BLUP, ssGBLUP, BayesB, and wssGBLUP were 0.19, 0.45, 0.61 and 0.63, respectively. The increased accuracy obtained by weighting SNP differently can be explained by presence of large QTL, by population structure (which can lead to fewer independent chromosome segments that may have larger effects), or by the small size of the genotyped population. In Manhattan plots, iteration number 3 had some SNPs explaining more than 7% of the genetic variance of the trait, meanwhile regular ssGBLUP had all regions explaining less than 0.25%. WssGBLUP utilizes all the available information, is simple to apply for complex models, and was the most accurate method in this study.