MegaLMM improves genomic predictions in new environments using environmental covariates.

Area of Science:

• Plant breeding
• Genetics
• Statistical modeling

Background:

• Multienvironment trials (METs) are essential for developing high-performing crop varieties.
• Current METs often lack sufficient environmental representation and struggle with climate change impacts.
• Accurate prediction of variety performance in new environments is critical for effective breeding programs.

Purpose of the Study:

• To extend the MegaLMM statistical model for genomic prediction in new environments.
• To leverage environmental covariates (ECs) to predict variety performance beyond existing METs.
• To assess the accuracy and utility of the extended MegaLMM in a large-scale maize dataset.

Main Methods:

• Developed an extended MegaLMM incorporating regressions of latent factor loadings on ECs.
• Utilized the maize Genome-To-Fields dataset (4,402 varieties, 195 trials, extensive missing data).
• Compared MegaLMM performance against univariate GBLUP for genomic prediction in new environments.

Main Results:

• Extended MegaLMM demonstrated high accuracy in genomic prediction across various breeding scenarios.
• MegaLMM significantly outperformed univariate GBLUP in predicting trait performance in novel environments.
• The study explored the use of higher-dimensional ECs for enhanced prediction accuracy.

Conclusions:

• The extended MegaLMM effectively enables genomic prediction in new environments using MET data and ECs.
• MegaLMM offers a powerful tool for plant breeding programs facing environmental variability and climate change.
• The methodology has broad applicability in genetics and breeding utilizing large-scale linear mixed models.