Optimizing Sparse Testing for Genomic Prediction of Plant Breeding Crops

Area of Science:

• Plant breeding and genetics
• Genomic selection
• Statistical genetics

Background:

• Genomic selection (GS) aims to enhance breeding program efficiency.
• Sparse testing strategies are proposed to reduce the number of genotypes evaluated per environment.
• Factors hindering the efficiency of sparse testing methods require investigation.

Purpose of the Study:

• To evaluate four sparse testing allocation methods (M1-M4) for multi-environmental trials.
• To assess the genomic prediction accuracy of unobserved lines using these methods.
• To compare uni-trait and multi-trait models within a sparse testing framework.

Main Methods:

• A two-stage analysis was employed to construct genomic training and testing sets.
• Sparse testing allowed each environment to evaluate only a subset of genotypes.
• Best Linear Unbiased Estimators (BLUEs) or Best Linear Unbiased Predictions (BLUPs) were computed in the first stage.

Main Results:

• The multi-trait model demonstrated superior genomic prediction (GP) accuracy compared to the uni-trait model.
• Allocation methods M3 and M4 slightly outperformed M1 and M2 in assigning lines to environments.
• Prediction accuracy remained high even with a 15-85% training-testing data split, showing minimal decrease.

Conclusions:

• Genomic sparse testing methods can significantly reduce operational and financial costs in breeding programs.
• These methods offer substantial resource savings with only a minor reduction in prediction precision.
• Sparse testing is a viable strategy for efficient genomic prediction in multi-environmental trials.