Data from: Genomic analysis and prediction within a US public collaborative winter wheat regional testing nursery

The development of inexpensive, whole-genome profiling enables a
transition to allele-based breeding using genomic prediction models. These
models consider alleles shared between lines to predict phenotypes and
select new lines based on estimated breeding values. This approach can
leverage highly-unbalanced datasets common to breeding programs. The
Southern Regional Performance Nursery (SRPN) is a public nursery
established by the USDA-ARS in 1931 to characterize performance and
quality of near-release wheat varieties from breeding programs in the US
Central Plains. New entries are submitted annually and can be reentered
only once. The trial is grown at more than 30 locations each year and
lines are evaluated for grain yield, disease resistance, and agronomic
traits. Overall genetic gain is measured across years by including common
check cultivars for comparison. We have generated whole-genome profiles
via genotyping-by-sequencing for 939 SPRN entries dating back to 1992. We
measured the diversity within the nursery and have explored its potential
use as a GS training population. GS prediction models across years
(average r= 0.33) outperformed year-to-year phenotypic correlation for
yield (r=0.27) for a majority of the years evaluated, suggesting that
genomic selection has the potential to outperform low heritability
selection on yield in these highly variable environments. We also examined
the predictability of programs using both program-specific and whole-set
training populations. Generally, the predictability of a program was
similar with both approaches. These results suggest that wheat breeding
programs can collaboratively leverage the immense datasets that are
generated from regional testing networks.