Dramatic reductions in cost of genotyping have now made it financially viable for genomic selection to become a routine tool in commercial wheat breeding. The work detailed here describes the process, challenges and impact of integrating genomic selection strategies into the wheat breeding programmes of Australian Grain Technologies. At AGT, Australia’s largest wheat breeding company, we have developed and used prediction calibrations for more than 30 traits. Training sets consist of up to 30,000 individuals and 230,000 phenotypic observations for yield, disease, and agronomic traits, and up to 6,000 individuals and 15,000 phenotypic observations for detailed end use quality.

Before genomic selection could be confidently utilised, several potentially derailing features of genomic prediction required further understanding. Extensive research on these issues was carried out, including previously published work by this group investigating issues such as linkage disequilibrium, population structure, prediction accuracy of qualitative and quantitative traits, and the effect of population size, relatedness, marker density and breeding cohort on genomic prediction accuracy with a training set of more than 10,000 individuals.

Here we present a summary of our experiences, including the application of genomic selection using “blind” calibrations, genomic selection borrowing from the power of sibling or family data, as well as the use of genomic selection indices to aid in optimal early generation selection.