Testing different single nucleotide polymorphism selection strategies for prediction of genomic breeding values in dairy cattle based on low density panels

Abstract

ABSTRACT:In human and animal genetics dense single nucleotide polymorphism (SNP) panels are widely used to describe genetic variation. In particular genomic selection in dairy cattle has become a routinely applied tool for prediction of additive genetic values of animals, especially of young selection candidates. The aim of the study was to investigate how well an additive genetic value can be predicted using various sets of approximately 3000 SNPs selected out of the 54 001 SNPs in an Illumina BovineSNP50 BeadChip high density panel. Effects of SNPs from the nine subsets of the 54 001 panel were estimated using a model with a random uncorrelated SNPs effect based on a training data set of 1216 Polish Holstein-Friesian bulls whose phenotypic records were approximated by deregressed estimated breeding values for milk, protein, and fat yields. Predictive ability of the low density panels was assessed using a validation data set of 622 bulls. Correlations between direct and conventional breeding values routinely estimated for the Polish population were similar across traits and clearly across sets of SNPs. For the training data set correlations varied between 0.94 and 0.98, for the validation data set between 0.25 and 0.46. The corresponding correlations estimated using the 54 001 panel were: 0.98 for the three traits (training), 0.98 (milk and fat yields, validation), and 0.97 (protein yield, validation). The optimal subset consisted of SNPs selected based on their highest effects for milk yield obtained from the evaluation of all 54 001 SNPs. A low density SNP panel allows for reasonably good prediction of future breeding values. Even though correlations between direct and conventional breeding values were moderate, for young selection candidates a low density panel is a better predictor than a commonly used average of parental breeding values.Keywords: 3K chip; genomic selection; prediction; single nucleotide polymorphism Since Meuwissen et al. (2001) proposed the application of genetic values predicted from a large number of single nucleotide polymorphisms (SNPs) for selection, many studies related to development of methodology and practical application of genomic selection have been conducted (for recent reviews see Hayes et al., 2009;Calus, 2010;Liu et al., 2011). Challenges remain, however: (i) in view of rapidly growing sizes of training data sets, how to deal with the large dimensions of the statistical model used for estimation of SNP effects; and (ii) in view of widespread genotyping of all members of active populations and all incoming selection candidates, how to reduce genotyping costs. While dimension reduction can be realized by the choice of an appropriate statistical model, a primary way of cost reduction is the application of cheaper, low density SNP panels. A currently widespread procedure in genomic evaluations is to use sparse commercially available 3K or 6K SNP panels geno-

Accuracy of genomic-polygenic estimated breeding value for milk yield and fat yield in the Thai multibreed dairy population with five single nucleotide polymorphism sets

“…The higher mean EBV accuracies obtained with four GG P80K subsets than with the complete SNP set supported the findings from previous research in dairy [5,27–29] and in beef cattle [30] that SNP subsets can yield comparable or higher levels of EBV accuracy than complete SNP sets while lowering genotyping costs.…”

Effect of DGAT1 polymorphisms on the estimated breeding values of Czech Simmental sires

“…Then, the genomic approach (Schopen et al 2009;Pribyl et al 2010Pribyl et al , 2012Pribyl et al , 2013Matejickova et al 2013;Szyda et al 2013) could be completed by the known major genes concerning milk performance and by the biometric approach (Sigl et al 2012;Meszaros et al 2013;Zavadilova and Stipkova 2013;Zavadilova and Zink 2013). When different effects depending on the breed are found, e.g.…”

Genomic prediction using preselected DNA variants from a GWAS with whole-genome sequence data in Holstein–Friesian cattle

“…However, when strongly selected variants were used, the genomic predictions became more biased, with slopes even lower than 0.5. Szyda et al [33] showed that genomic predictions for milk yield were biased when 3 k variants were selected for their effect on milk yield [33]. Brondum et al [32] reported no extra bias when 1623 selected SNPs were added to the 54 k SNP panel, but when the 1623 SNPs were fitted with their own variance in a model with the 54 k SNPs, the bias increased for some traits.…”