Genomic Prediction Enhanced Sparse Testing for Multi-environment Trials.

Area of Science:

• Plant breeding and genetics
• Agricultural science
• Statistical genomics

Background:

• Sparse testing in multi-environment breeding trials reduces costs and increases capacity by testing fewer genotypes per environment.
• Genomic-enabled prediction models can predict unobserved genotype-in-environment combinations, but accuracy depends on overlap, testing frequency, and prediction method.
• Optimizing sparse testing designs is crucial for efficient crop improvement.

Purpose of the Study:

• To evaluate the predictive ability of different sparse testing designs using genomic-enabled prediction models.
• To compare the performance of models with and without genotype × environment (GE) interaction.
• To determine the impact of genotype overlap and calibration set size on prediction accuracy.

Main Methods:

• Utilized two maize hybrid datasets with phenotypic yield data from three environments.
• Implemented three prediction models: two main effects models (M1, M2) and one including GE interaction (M3).
• Assessed various sparse testing designs, from no genotype overlap to complete overlap between environments.

Main Results:

• The genome-based model including GE (M3) captured more phenotypic variation and showed higher prediction accuracy than M1 and M2.
• Reducing calibration set size decreased prediction accuracy, but M3 was less affected.
• Genomic-enabled models (M2, M3) recovered predictive ability when more genotypes were tested across environments.

Conclusions:

• Genome-based models incorporating GE are effective for optimizing sparse testing designs in plant breeding.
• Substantial savings in testing resources are achievable without compromising predictive accuracy.
• Strategic design of sparse testing, informed by genomic prediction, enhances breeding program efficiency.