PSX-42 Assessing Haplotype-based Genomic Predictions in Canadian Swine Breeds.

Abstract

Individual single nucleotide polymorphism (SNP) genotypes are currently used to build the genomic relationship matrix (G) for genomic best linear unbiased prediction (GBLUP) in Canada. While SNPs are bi-allelic, haplotypes are usually “multi-allelic” and, thus, they may better capture multi-allelic quantitative trait loci (QTL) compared to individual SNPs. The aims of this study were to compare the performance of haplotype-based GBLUP (G_HAPBLUP) to individual SNP-based GBLUP (G_SNPBLUP) with respect to prediction reliability and prediction bias in 3 major Canadian swine breeds, i.e., Duroc, Landrace, and Yorkshire. Genomic relationship matrices based on individual SNP genotypes or haplotype blocks of different sizes (i.e., 5, 10, 15, 20 and 30) were built using imputed 70K SNP genotypes and, then, they were used to predict genomic estimated breeding values (GEBVs) for backfat thickness, total number of born, age adjusted to 100kg of live weight, and loin depth adjusted to 100kg of live weight. In addition, Spearman rank correlation between G_SNPBLUP and G_HAPBLUP GEBVs was calculated to evaluate how strongly the use of haplotypes instead of individual SNPs change the ranking of selection candidates. Except for backfat thickness in Duroc, G_HAPBLUP increased prediction reliabilities (from +0.3 to +8.2% points) and reduced prediction biases (from +0.026 to +0.112 points), respectively, across the breeds and traits studied. Furthermore, the optimum haplotype block size was not consistent across traits and breeds. However, longer haplotype blocks (i.e., 15, 20 and 30) resulted in higher improvements in prediction reliabilities and less prediction biases compared to shorter haplotypes (i.e., 5 and 10). Moreover, the Spearman rank correlations indicated that G_HAPBLUP leads to substantial re-ranking of top selection candidates, thus potentially affecting selection decisions in swine breeding programs. In conclusion, G_HAPBLUP outperformed G_SNPBLUP in terms of predictive ability and it is an alternative to G_SNPBLUP for genomic evaluation in swine breeding.