Genomic prediction in multi-environment trials in maize using statistical and machine learning methods.

Area of Science:

• Agricultural Science
• Plant Breeding
• Genetics

Background:

• Genomic prediction offers a cost-effective alternative to traditional field trials for evaluating untested single cross hybrids in multi-environment trials (MET).
• Accurate prediction of hybrid performance is crucial for accelerating maize breeding programs.

Purpose of the Study:

• To evaluate genomic prediction for grain yield and female flowering time in untested single cross maize hybrids within MET.
• To explore and compare machine learning methodologies against Genomic Best Linear Unbiased Prediction (GBLUP) with non-additive effects for hybrid prediction in MET.

Main Methods:

• Genomic prediction models were applied to predict phenotypes of single cross hybrids not included in field trials.
• Machine learning approaches were investigated and compared with GBLUP, considering non-additive genetic effects.

Main Results:

• Both machine learning and GBLUP demonstrated efficiency in predicting hybrid performance across different environments.
• Predicting entirely novel hybrids was more challenging than predicting hybrids in sparse test designs.
• The optimal prediction methodology is context-dependent, requiring careful variance component modeling for GBLUP or leveraging machine learning's ability to capture non-additive effects without prior assumptions.

Conclusions:

• Genomic prediction, utilizing both GBLUP and machine learning, is a valuable tool for maize breeding programs.
• Accurate modeling of variance components is key for optimizing GBLUP, while machine learning offers flexibility in capturing complex genetic interactions.