Experiences with a single-step genome evaluation

Abstract

Genomic selection can be implemented based on the genomic relationship matrix (GBLUP) and can be combined with phenotypes from nongenotyped animals through the use of best linear unbiased prediction (BLUP). A common method to combine both sources of information involves multiple steps, but is difficult to use with complicated models and is nonoptimal. A simpler method, termed single-step GBLUP, or ssGBLUP, integrates the genomically derived relationships (G) with population-based pedigree relationships (A) into a combined relationship matrix (H) and allows for genomic selection in a single step. The ssGBLUP method is easy to implement and uses standard BLUP-based programs. Experiences with field data in chickens, pigs, and dairy indicate that ssGBLUP is more accurate yet much simpler than multi-step methods. The current limits of ssGBLUP are approximately 100,000 genotypes and 18 traits. Models involving 10 million animals have been run successfully. The inverse of H can also be used in existing programs for parameter estimationm, but a properly scaled G is needed for unbiased estimation. Also, as genomic predictions can be converted to SNP effects, ssGBLUP is useful for genomic-wide association studies. The single-step method for genomic selection translates the use of genomic information into standard BLUP, and variance-component estimation programs become a routine.