An Evaluation of 'Deflation' to Improve Convergence Rates for Single-Step Genomic Evaluation with the Hybrid Model

Single step genomic evaluation fitting a 'hybrid' model which combines marker effects for individuals with genotypes with breeding values for non-genotyped animals can readily accommodate large numbers of genotyped animals. However, iterative solution of the pertaining mixed model equations via a preconditioned gradient scheme has been reported to be afflicted by much slower convergence rates than the standard breeding value model. 'Deflation' of the coefficient matrix has been proposed as a second preconditioning step and shown to dramatically reduce numbers of iterations and computing time required. We describe its application for a set of sheep data. Results indicate that assignment of marker effects to subdomains in moderately sized chunks together with a separate treatment of genetic group effects could reduce total computing times by about a third.