Genetic evaluation using unsymmetric single step genomic methodology with large number of genotypes

Abstract

The single step genomic methodology provides a unified framework to integrate phenotypic, pedigree and genomic information in the prediction of breeding values. Minimal modifications of current softwares are necessary in order to incorporate extra relationship matrices, however computing such matrices has a cubic cost. Recently, a system of equations relaxing the computing cost of creating the inverse of the genomic relationship matrix was presented, which creates an unsymmetric system of equations. Bi Conjugate Gradient Stabilized solvers (BiCGSTAB) were proposed to solve unsymmetric system of equations and also can be used with iteration on data programs, resulting in a good choice for solving large-scale genetic evaluations. Here we describe the implementation of a large genetic evaluation using unsymmetric solvers within the iteration on data framework. Comparison with the regular single-step methodology is presented and the effects of different preconditioners and data structures on the convergence pattern were studied. A large scale genetic evaluation was feasible, however required more rounds to get convergence compared with the regular single-step. More sophisticated preconditioners are necessary to improve the convergence for solving unsymmetric single-step genomic evaluations.