Responses of imputation accuracy on marker density and individual relationship*.

*The full dataset from 15th QTL-MAS workshop was sampled on individual relationship and marker density. The full dataset contains 3220 individuals genotyped with 9990 markers. The 3220 individual include 20 sires, 200 dams (10 dam per sire), and 3000 progeny (15 progeny per dam) as displayed in Figure S1a. The full population were randomly sampled to form two sub populations, one with individuals more related each other (full sibs see Figure S1c) and the other with individuals less related each other (half sibs, see Figure S1b). The known genotypes were randomly masked as missing at three different rates: 60%, 70%, and 80%. Two imputation methods (BEAGLE, fastPHASE and iBLUP) were used to impute the masked genotypes. Accuracy was calculated as Pearson correlation coefficient between known genotype and imputed. The sampling of missing genotypes was repeated ten times. The average and standard error of imputation accuracy are reported in the table.aAll the genetic markers were used to evaluate the responses of imputation accuracy on individual relationship, i.e. half sib vs. full sibs population.bThe half sib population was used to evaluate the responses of imputation accuracy on marker density. Two levels of marker density were examined. The high level marker density contained all the available markers (9990). The low density contained one fifth of the total available markers which are sampled evenly (choosing one out of every five adjacent markers).