Abstract: Genomic selection has been demonstrated as a powerful technology to revolutionize animal breeding.  However, marker density and minor allele frequency can affect the predictive ability of genomic estimated breeding values (GEBVs).  To investigate the impact of marker density and minor allele frequency on predictive ability, we estimated GEBVs by constructing the different subsets of single nucleotide polymorphisms (SNPs) based on varying markers densities and minor allele frequency (MAF) for average daily gain (ADG), live weight (LW) and carcass weight (CW) in 1 059 Chinese Simmental beef cattle.  Two strategies were proposed for SNP selection to construct different marker densities: 1) select evenly-spaced SNPs (Strategy 1), and 2) select SNPs with large effects estimated from BayesB (Strategy 2).  Furthermore, predictive ability was assessed in terms of the correlation between predicted genomic values and corrected phenotypes from 10-fold cross-validation.  Predictive ability for ADG, LW and CW using autosomal SNPs were 0.13±0.002, 0.21±0.003 and 0.25±0.003, respectively.  In our study, the predictive ability increased dramatically as more SNPs were included in analysis until 200K for Strategy 1.  Under Strategy 2, we found the predictive ability slightly increased when marker densities increased from 5K to 20K, which indicated the predictive ability of 20K (3% of 770K) SNPs with large effects was equal to the predictive ability of using all SNPs.  For different MAF bins, we obtained the highest predictive ability for three traits with MAF bin 0.01–0.1.  Our result suggested that designing a low-density chip by selecting low frequency markers with large SNP effects sizes should be helpful for commercial application in Chinese Simmental cattle.

Key words: genomic prediction, cross-validation, Chinese Simmental beef cattle, marker density, minor allele frequency (MAF)