A simulation tool to support the implementation of genomic selection in the UK guide dogs’ population

In the last two decades, genomic breakthroughs have allowed major breeding and genetic improvements for the health and productivity of many animal populations. In dogs, the knowledge and testing of genetic diseases have reduced their incidence in major breeds. However, the population structure did not allow the implementation of an advanced breeding program. For service dog populations, applying genomic selection would enable more efficient breeding for health, welfare, and trainability. However, the transition between phenotypic to genomic selection requires genomic information for a large part of the population. Different data collection scenarios can be envisioned based on the number and type of individuals, the number of markers, and the genotyping technology. These criteria affect the yielded data accuracy and its cost. The goal of this study was to create a simulation to evaluate genotyping scenarios to find the best strategy for the implementation of genomic selection to improve the Guide Dogs’ population. To do so, we used AlphaSimR, an R package for stochastic simulations of breeding programs down to the DNA level, and AlphaPeel, a software for calling, phasing, and imputing genotypes. First, genomes of the founders were simulated based on the characteristics of the organisation’s breeding programme. Then, the pedigree was used to drop founder genomes to the rest of the population via a known pedigree. This approach also enables creation of new descendants by extending the pedigree with the simulation of the population mating structure. Simulated true genotypes were available for the whole pedigree and used to generate scenario-specific genetic datasets with SNP array genotypes or whole-genome sequences at variable sequencing depth. AlphaPeel then propagated information from the genetic datasets for genotyped individuals to the ungenotyped members of the population through their pedigree connections. Such scenarios can be ranked based on simulated accuracy of imputation and cost. The advantage of this simulation is in its adaptability. For example, available phased SNP array genotypes or whole-genome sequences can be integrated to obtain a more realistic simulation. Provided biological parameters as priors, the simulation can be fitted to most populations and can be widely used for hypothesis testing and optimisation. In conclusion, simulation is a powerful and flexible tool used for hypothesis testing, applied here to determine the best genotyping strategy in a guide dog population.