Investigating the relationship between imputation accuracies and relatedness

Single Step Best Linear Unbiased Prediction (ssBLUP) is used in the Australian beef industry's genetic evaluation, BREEDPLAN, for the prediction of Estimated Breeding Values (EBVs), which uses genomic information in a Genomic Relationship Matrix (GRM). Imputation of missing Single Nucleotide Polymorphisms (SNPs) and imputation of low density to high density genotypes is essential to combine various SNP densities for building the GRM. EBV accuracies are dependent on an individual's relationship to the rest of the population. Similarly, a target population's imputation accuracy is dependent on relatedness to the reference population. This study introduces a 'relatedness score', calculated in a similar way to EBV accuracies, to indicate an animal's relatedness to the reference population. The objective of this study was to identify how well the relatedness score can predict imputation accuracies. For this purpose QMSim was used to simulate 10 generations of genotypes (20 chromosomes and 2000 SNPs – 200 cM) with 40 males and 800 females in the historical population. Generations 4 to 10 were used to evaluate the relationship between imputation accuracies and relatedness score. The results demonstrated a non-linear correlation between imputation accuracies and relatedness score when individuals exist across multiple generations and with densities greater than 1000 SNPs were used. These results indicate a relatedness score may explain low EBV accuracies and EBV instability within BREEDPLAN data, due to low imputation accuracies.