Genomic prediction using multi-year data: A case study in a hybrid maize breeding program

Abstract

Single-cross hybrids have been used to explore heterosis in selfing and outcrossing species. However, as it is infeasible to obtain and evaluate all possible combinations among pairs of inbred lines, predict the performance of untested single-cross hybrids is essential to increase genetic gains in breeding programs. In a breeding routine, data are often unbalanced because different sets of hybrids are phenotyped in different selection cycles (years) and in multi-environment trails. In this context, appropriate statistical models that account for genetic and residual correlations across environments, years and deals with unbalanced data is required. Moreover, in species with high level of heterosis, such as maize, it is appropriate that the genomic selection models consider not only additive genetic effects, but also the non-additive genetic effects (dominance and epistatic) to the prediction the performance of untested hybrids. Thus, this project goal is to evaluate the predictive accuracy within and across breeding cycles of single-cross maize hybrids for grain yield. Data of 748 hybrids from different breeding cycles evaluated from 2006 to 2013 will be used to asses the potential of prediction performance. Genotypic data are available via genotyping-by-sequencing for the inbred lines used as parents of the evaluated hybrids. For this, a statistical-genetics model that accounts for genotype-by-environment and genotype-by-year interaction, additive and dominance effects will be proposed. Then, in addition to the practical and theoretical results applied to the maize hybrid breeding program, the conclusions achieved in this project may be applied for any crop in which hybrids are used to explore heterosis. Likewise, this approach has potential to reduce costs and accelerate the release of new hybrids.