Effect of reference population size and available ancestor genotypes on imputation of Mexican Holstein genotypes

The effects of reference population size and the availability of information from genotyped ancestors on the accuracy of imputation of single nucleotide polymorphisms (SNP) were investigated for Mexican Holstein cattle. Three scenarios for reference population size were examined: (1) a local population of 2,011 genotyped Mexican Holsteins, (2) animals in scenario 1 plus 866 Holsteins in the US genotype database (GDB) with genotyped Mexican daughters, and (3) animals in scenario 1 and all US GDB Holsteins (338,073). Genotypes from 4 chip densities (2 low density, 1 mid density, and 1 high density) were imputed using findhap (version 3) to the 45,195 markers on the mid-density chip. Imputation success was determined by comparing the numbers of SNP with 1 or 2 alleles missing and the numbers of differently predicted SNP (conflicts) among the 3 scenarios. Imputation accuracy improved as chip density and numbers of genotyped ancestors increased, and the percentage of SNP with 1 missing allele was greater than that for 2 missing alleles for all scenarios. The largest numbers of conflicts were found between scenarios 1 and 3. The inclusion of information from direct ancestors (dam or sire) with US GDB genotypes in the imputation of Mexican Holstein genotypes increased imputation accuracy by 1 percentage point for low-density genotypes and by 0.5 percentage points for high-density genotypes, which was about half the gain found with information from all US GDB Holsteins. A larger reference population and the availability of genotyped ancestors improved imputation; animals with genotyped parents in a large reference population had higher imputation accuracy than those with no or few genotyped relatives in a small reference population. For small local populations, including genotypes from other related populations can aid in improving imputation accuracy.