Comparison of genomic prediction models for general combining ability in early stages of hybrid breeding programs

Abstract This study evaluates the impact of genomic prediction models on selecting inbred lines as parents in hybrid breeding programs. New parents in a hybrid breeding program are typically selected from early‐stage yield trials based on general combining ability (GCA) from testcrosses. Genomic studies have largely focused on predicting hybrid performance in the late stages of the breeding pipeline and largely ignored the selection of inbred lines as parents of the subsequent breeding cycles. Here, we used stochastic simulations of a maize ( Zea mays L.) hybrid breeding program for 20 years to evaluate the performance of genomic prediction models for selecting parents based on their predicted GCA. Five genomic prediction models were evaluated in terms of achieved genetic gain and heterosis under two different single nucleotide polymorphism (SNP) marker densities and the true quantitative trait loci genotypes. The results show that using high‐density SNP markers generated more genetic gain and heterosis than the low‐density SNP markers. The relative performance of genomic prediction models differed across marker scenarios. For genetic gain, we observed more differences between the models at low than high marker density. For heterosis, we observed the opposite, more differences between the models at high than low marker density. Overall, models that fitted the average or additive effects specific to each heterotic pool and dominance effects provide a better fit and hence higher genetic gain in hybrid breeding programs.