关键词: 填充准确性, 低密度SNP芯片, 华西牛, 连锁不平衡, 最小等位基因频率

Abstract: 【Objective】To facilitate low-cost genomic selection in Huaxi Cattle, the present study represent the first attempt to designed a new low-denstity Genotype chip to support imputation to higher density genotypes. The representative SNP markers with different density gradients were selected from high-density SNP chips in the Huaxi Cattle reference population by using two SNP selection methods. And then these marker sets were imputed to high-density sets with the same imputation parameters for subsequent genomic studies. Meanwhile, the current study compared the differences in imputation accuracy and concordance among SNP panels and illustrated the effects of four factors on imputation results including marker screening method, marker density, minor allele frequency and the number of reference population. This study provides insights about the methods to select low-density SNP markers for imputation in the current population and the representative SNPs will aid in designing low-density SNP chip for Huaxi cattle.【Method】Totally 1,233 Huaxi cattle after genotypes filtered was randomly divided into reference (986) and validation (247) populations. two SNP selection strategies, based on Equidistance (EQ) and on high MAF (HM), were used to make 16 SNP sets with different densities from the Illumina Bovine HD chip in the reference population, respectively. Each of the 32 low-density set was then imputed to the 770K density level in the validation population using Beagle (v5.1), while the imputation accuracy and concordance were calculated as the mean correlation between true and imputed genotypes. While, a comprehensive set of factors that influence the imputation performance were analyzed.【Result】The number of markers in the 32 low-density SNP sets ranged from 100 to 16 000, with a maximum window of 24 176 kb and a minimum window of 151 kb. The imputation accuracy and concordance of both EQ and HM methods went up with increasing marker densities. The imputation accuracy of both methods was highest at 16k SNP density （r²_EQ=0.8801，r²_MAF=0.8696）. When the marker density was below 11k, the imputation concordance of HM was higher than EQ for all marker density gradients. However, when the SNP density exceeded 11k, EQ showed an imputation accuracy advantage over HM. Similar to the imputation concordance results, the HM method still had higher imputation accuracy when the SNP density was lower than 10k, but the EQ method had higher imputation accuracy when the SNP pool density was higher than 10k, and the EQ imputation accuracy tended to be stable after the SNP density was greater than 12k. It was also found that the imputation accuracy of high MAF locus was higher. During the imputation process, it was found that the imputation accuracy and concordance increased with the increase of the reference panel. The imputation accuracy and concordance of loci were higher when the population of the reference panel was 600-800. 【Conclusion】In the Huaxi cattle population, imputation accuracy and concordance increased with increasing marker density, and a better imputation effect could be obtained in the marker density of 10k-12k interval. The HM method was preferred when the marker density was less than 10k, and the EQ method was better at high marker density. High MAF locuses were more accurate for imputation. When using the imputation strategy for low-density marker imputation, the number of reference panel should be at least 400 heads for better imputation effect.

Key words: imputation accuracy, low density SNP array, Chinese Simmental cattle, linkage disequilibrium, MAF