Genomic prediction models for traits differing in heritability for soybean, rice, and maize

Area of Science:

• Plant breeding
• Genomics
• Quantitative genetics

Background:

• Genomic selection (GS) utilizes marker and phenotype data to predict breeding values.
• Optimizing prediction accuracy in GS models is crucial for effective plant breeding.
• Cross-species evaluation of GS models is needed due to variations in linkage disequilibrium (LD) and genetic architecture.

Purpose of the Study:

• To evaluate 11 genomic prediction models for prediction accuracy across maize, soybean, and rice.
• To assess the impact of marker subsets and training population size on prediction accuracy.
• To identify optimal strategies for enhancing genomic prediction accuracy.

Main Methods:

• Cross-validation was used to assess prediction accuracy.
• Tested three training-to-testing population proportions (90:10, 70:30, 50:50).
• Evaluated marker subsets selected by LD and statistical significance (P≤0.05 or P≤0.10).

Main Results:

• Maize, with the shortest LD, exhibited the highest prediction accuracy.
• The Bayes B model consistently outperformed other models across species and traits.
• Higher heritability traits showed greater prediction accuracy.
• Using significant markers improved accuracy compared to complete marker sets; excluding QTL-associated markers reduced accuracy.

Conclusions:

• The Bayes B model combined with significant marker subsets enhances genomic prediction accuracy.
• Training population selection based on narrow-sense heritability is recommended.
• These findings provide strategies for optimizing genomic selection in diverse plant species.