Impact of an Approximate Inverse of the Genomic Relationship Matrix for Single-Step Evaluation of Australian Meat Sheep

Common implementations of single-step genomic evaluation require the inverse of the genomic relationship matrix. Obtaining the inverse can become computationally prohibitive as its size increases. Stimulated by rapidly increasing numbers of genotyped animals, several procedures to approximate this inverse have been proposed. We examine the impact of two methods of approximation on predicted breeding values for a multi-breed population of Australian sheep. Results show that very high correlations with predictions using the full inverse can be achieved whilst reducing computational requirements. However, current levels of genotyping in our data were relatively low and results need to be validated as larger number of genotypes become available.