Impact of the Genetic Diversity of the Reference Population on the Accuracy of Imputation of Rare Variants

The cost for whole-genome sequencing is still too high for use on large number of animals. SNP genotyping with chips is routinely used and common cattle breeds are widely genotyped. Imputation algorithms have been developed to allow for in silico inference of whole-genome sequence (WGS) for all genotyped individuals. High average imputation accuracies are commonly observed. Accurate imputation of rare variants, however, is still difficult and strongly dependent on the composition of the reference population. Up to now most of the sequenced cattle individuals have been selected using the key ancestor approach. This method relies on pedigree or genomic relationships and aims to optimize the genotypic variance of a population explained by the selected animals. In order to reach better imputation accuracies of rare variants in a genotyped population, we have developed a new algorithm for choosing the candidates for sequencing. Following a simulated annealing iteration method, we optimize the Genetic Diversity Index (GDI) of a group of animals that will be the future reference population for imputation. This GDI is computed by summing the count of unique haplotypes at each 20-SNP-window over all animals present in a selection. Here we will simulate a set of WGS using coalescence and gene dropping along a North-American Holstein pedigree. WGS will be masked to array density and imputed back to WGS to compare accuracies using differently selected reference groups. We expect that our approach will yield higher imputation accuracy of rare variants, relying on a more genetically diverse reference population.