Genomic predictability of interconnected biparental maize populations.

Area of Science:

• Plant breeding and genetics
• Quantitative genetics
• Genomic selection

Background:

• Designing effective training sets (TS) is critical for genomic selection (GS) in structured plant breeding populations.
• Understanding how to best construct a TS from multiple related or unrelated biparental families to predict progeny from individual crosses remains an open question.

Purpose of the Study:

• To systematically investigate the impact of training set composition on prediction accuracy in genomic selection.
• To evaluate how different types of relatedness (full-sibs, half-sibs, unrelated) within the TS influence prediction of progeny from individual crosses in maize.

Main Methods:

• Utilized five interconnected maize (Zea mays L.) doubled-haploid (DH) line populations derived from four parents.
• Genotyped 635 DH lines with 16,741 polymorphic SNPs and evaluated five traits, including Gibberella ear rot and kernel yield components.
• Analyzed genomic similarity patterns to define relationships (full-sibs, half-sibs, unrelated) and assessed prediction accuracies based on TS composition.

Main Results:

• Prediction accuracies within full-sib families closely matched theoretical expectations.
• Replacing full-sibs with half-sibs in the TS decreased prediction accuracies by 42%, but using half-sibs from both parents of the validation population improved results.
• Including unrelated crosses with opposite linkage phases negatively impacted prediction accuracy, suggesting their exclusion from the TS.

Conclusions:

• The genomic relatedness of lines within the training set significantly influences prediction accuracy in genomic selection.
• Prioritizing related individuals, particularly full-sibs, and excluding unrelated lines with unfavorable linkage phases are recommended for optimizing TS design.
• Trait and population variability necessitate adaptive strategies for resource allocation in genomic selection model development.