The effect of genotype imputation and some important factors on the accuracy of genomic prediction and its persistency over time

Abstract

This study was conducted to investigate the effect of genotype imputation on accuracy of genomic prediction. In addition, the effect of heritability, marker density, and reference population size on the accuracy of genomic prediction and its persistency over time was also studied. To this end, a genome comprised of 3 chromosomes on which 6000 bi-allelic SNP with equal initial frequency of 0.5 were arrayed was simulated. For assessing the genotype imputation, respectively, 5%, 30%, 50%, 70%, 90% and 95% of genotypes were masked and then imputed with Random Forest algorithm. Two reference population size of 1000 and 5000 animals and two marker densities of 20 SNPs and 100 SNPs per centimorgan were considered. Correlation coefficient between the true and predicted genomic breeding values (rp,t) was used as the prediction accuracy. The imputation accuracy (IA) decreased in parallel to increase in the percentage of imputed genotypes. It was ≈1 when only 5% of genotypes were imputed; however, as the proportion of masked genotypes increased beyond 70%, a rapid decline in IA was observed. In addition, by increasing the percentage of masked genotypes, rp,t decreased in such a way that it decreased from 0.78 to, respectively, 0.77, 0.75, 0.73, 0.70, 0.61, and 0.51 when, respectively, 5%, 30%, 50%, 70%, 90% and 95% of genotypes were imputed. Therefore, genotype imputation could be used effectively to decrease the cost of genomic selection. Significant increases were observed in accuracy of genomic prediction following increase in heritability, marker density and reference population size. With a heritability level of 0.2, marker density of 20 SNPs/cM and a reference population size of 1000 individuals, after 9 generations derived from the reference population, the accuracy decreased by 26% (0.78 to 0.57). By increasing marker density to 100 SNP/cM, the rp,t decreased by 16% (from 0.88 to 0.72). Moreover, by increasing reference population size from 1000 to 5000 individuals, the accuracy decreased by 14% (from 0.92 to 0.79) after 9 generations. In addition, by increasing heritability from 0.2 to 0.5, after 9 generations, there was 14% decrease in rp,t. It was concluded that increasing the reference population size and marker density could be effective strategies to prevent decrease in accuracy of genomic evaluation over time.