Constructing training sets for genomic selection to identify superior genotypes in candidate populations

Area of Science:

• Plant breeding
• Genomics
• Quantitative genetics

Background:

• Identifying superior genotypes is crucial for plant breeding programs.
• Genomic selection (GS) aims to enhance breeding efficiency through genomic information.
• Optimizing training set construction is key to improving GS performance.

Purpose of the Study:

• To evaluate and propose methods for training set optimization in genomic selection.
• To enhance the efficiency of identifying top-performing genotypes from breeding populations.
• To compare optimal design-inspired approaches with accuracy-focused methods.

Main Methods:

• Evaluation of various training set construction methods for genomic selection.
• Implementation of two novel approaches inspired by optimal design criteria.
• Comparison using simulation studies and real trait data analysis.
• Assessment using metrics like normalized discounted cumulative gain and correlation coefficients.

Main Results:

• A ridge regression-based method is recommended for populations without strong subpopulation structure.
• Heuristic-based generalized coefficient of determination and D-optimality-like methods are preferred for structured populations.
• A ranking method can down-scale large candidate populations before applying heuristic methods.
• A computationally efficient version of a known method was verified.

Conclusions:

• The choice of training set construction method in genomic selection depends on population structure.
• Optimized training sets improve the efficiency of identifying superior genotypes in plant breeding.
• Proposed methods offer effective strategies for enhancing genomic selection performance.