Enhancing Across-Population Genomic Prediction for Maize Hybrids

Area of Science:

• Plant breeding and genetics
• Genomics
• Quantitative genetics

Background:

• Genomic selection (GS) uses genome-wide markers to predict phenotypes and enhance genetic gain in crop breeding.
• Practical GS applications often face challenges with predictions across genetically dissimilar populations.
• Developing effective strategies for across-population prediction is crucial for broader GS implementation.

Purpose of the Study:

• To evaluate the impact of training population optimization on across-population prediction accuracy in maize.
• To develop and assess a novel population structure-based across-population genomic prediction (PSAPGP) strategy.

Main Methods:

• Utilized a large maize training population (5820 hybrids) to predict a smaller breeding population (523 hybrids) using GBLUP and BayesB models.
• Optimized training population size and genetic relationships to assess prediction accuracy.
• Implemented PSAPGP by incorporating population structure (Q-matrix) as a fixed effect in GS models, assessed via PCA, clustering, and Q-matrix analysis.

Main Results:

• Prediction accuracy showed improvement with optimized training population sizes, though optimal sizes varied by trait.
• The PSAPGP strategy, particularly using the Q-matrix, significantly enhanced across-population prediction performance.
• Improvements of 8-11% in prediction accuracy were observed for ear weight, ear grain weight, and plant height using the PSAPGP strategy.

Conclusions:

• Optimizing training population size and genetic relatedness can improve across-population prediction accuracy.
• The proposed PSAPGP strategy offers a promising approach for accurate genomic prediction across diverse populations.
• This strategy has the potential to reduce phenotyping costs and accelerate maize hybrid breeding programs.