The goal of this study was to compare significant SNP selection approaches in the context of complex traits based on SNP estimates obtained by models: a model fitting a single SNP (M1), a model fitting a single SNP and a random polygenic effect (M2), the nonparametric CAR score (M3), a SNP-BLUP model with random effects of all SNPs fitted simultaneously (M4). There were 46,267 SNPs tested in a population of 2601 Holstein Friesian bulls, four traits (milk and fat yields, somatic cell score, non-return rate for heifers) were considered. The numbers of SNPs selected as significant differed among models. M1 selected a very large number of SNPs, except for a NRH in which no SNPs were significant. M2 and M3 both selected similar and low number of SNPs for each trait. M4 selected more SNPs than M2 and M3. Considering linkage disequilibrium between SNPs, for MY M2 and M3 selected SNPs more highly correlated with each other than in the case of M4, while for FY M3 selection contained more correlated SNPs than M2 and M4. In conclusion, if the research interest is to identify SNPs not only with strong, but also with moderate effects on a complex trait a multiple-SNP model is recommended. Such models are capable of accounting for at least a part of linkage disequilibrium between SNPs through the design matrix of SNP effects. Functional annotation of SNPs significant in M4 reveals good correspondence between selected polymorphisms and functional information as well as with QTL mapping results.
Keywords: Complex traits; GWAS; Mixed model; Significance testing.