Multi-Trait Genomic Selection in Plant Breeding Programs

Predictive models that leverage genomic variability allow crop improvement programs to make large gains in efficiency and logistical resources via virtual breeding. In nearly all breeding programs there are multiple traits of interest which can be at cross-purposes. For example, when selecting potential crosses a breeder should have the responsibility and capability to not only drive gains in yield and disease resistance, but also in vitamin content and production ease to improve global human health. In this presentation, we will focus on prediction methods such as RRBLUP (ridge regression) and GBLUP (Genomic best linear unbiased predictor) utilized in a cross-validated model comparison scheme to compare and trade-off multiple traits simultaneously. Model scoring is used to perform analytic cross-evaluation and selection based on multi-trait optimization, and then combined with progeny simulation to execute a multi-year in-silico breeding program that allows breeders to assess thousands of potential crosses that would not be feasible in the field. Corn breeding data provided by the Brazilian Agricultural Research Corporation (Embrapa) will be used as demonstration.