Genomic prediction in hybrid breeding: I. Optimizing the training set design

Area of Science:

• Plant breeding
• Quantitative genetics
• Genomic prediction

Background:

• Genomic prediction (GP) is vital for accelerating hybrid breeding.
• The optimal composition of training sets (TS) for GP, specifically the number of parents (n_TS) and crosses per parent (c), remains under-explored.
• Understanding how TS structure impacts prediction accuracy for different hybrid generations (H0, H1, H2) and parental lines (I0, I1) is essential.

Purpose of the Study:

• To investigate the impact of training set composition on genomic prediction accuracy.
• To evaluate prediction accuracy for General Combining Ability (GCA) of parental lines and hybrid performance (H0, H1, H2) under varying TS designs.
• To develop and validate theoretical estimates for prediction accuracy in hybrid breeding scenarios.

Main Methods:

• Developed theoretical estimates for prediction accuracy of GP using GBLUP (Genomic Best Linear Unbiased Prediction).
• Simulated hybrid populations using molecular data from maize, incorporating additive and dominance effects of quantitative trait loci (QTL).
• Assessed prediction accuracy across six scenarios with varying specific combining ability (SCA) variance proportions and heritability.

Main Results:

• Theoretical estimates ([Formula: see text] and [Formula: see text]) closely matched simulation-based prediction accuracy ([Formula: see text]) for hybrids.
• Training sets maximizing parent lines per cross (c=1) yielded highest accuracy for H0 hybrids and I0 lines.
• Conversely, c=1 resulted in the lowest prediction accuracy for I1 lines and H1/H2 hybrids across all scenarios and datasets.

Conclusions:

• The optimal training set composition for genomic prediction in hybrid breeding is dependent on the target trait and hybrid generation.
• A strategy maximizing parent lines per cross benefits early-generation hybrids (H0), while other strategies are needed for later generations (H1, H2) and parental lines.
• Breeding programs must consider their specific resources and objectives when designing training sets to maximize selection response across all hybrid types.