Multi-Trait Multi-Environment Genomic Prediction for End-Use Quality Traits in Winter Wheat

Area of Science:

• * Agricultural Science
• * Plant Breeding
• * Genetics

Background:

• * Soft white wheat quality phenotyping is delayed due to cost and time constraints.
• * Genomic selection (GS) offers potential for earlier genotype selection in wheat breeding.
• * Multi-trait GS models can leverage information from various traits and locations.

Purpose of the Study:

• * To evaluate multi-trait genomic selection models for predicting seven end-use quality traits in soft white wheat.
• * To compare the performance of uni-trait and multi-trait GS models.
• * To assess prediction accuracy across different environments and prediction scenarios.

Main Methods:

• * Utilized a population of 666 soft white wheat genotypes grown over 5 years in two Washington locations.
• * Optimized and compared four GS models: Bayes B, genomic best linear unbiased prediction (GBLUP), multilayer perceptron (MLP), and random forests.
• * Employed cross-validation, independent prediction, and across-location prediction strategies.

Main Results:

• * Multi-trait GS models showed 5.5% and 7.9% superior prediction accuracy compared to uni-trait models for within-environment and across-location predictions, respectively.
• * Multi-trait machine and deep learning models outperformed GBLUP and Bayes B for across-location predictions, with diminished advantage when genotype-by-environment interaction was included.
• * The multi-trait MLP model achieved a 35% improvement in prediction accuracy for flour protein content.

Conclusions:

• * Multi-trait GS models significantly enhance prediction accuracy for soft white wheat quality traits.
• * These models can accelerate the wheat breeding cycle in a cost-effective manner.
• * Leveraging previously phenotyped traits through multi-trait GS is a promising strategy for crop improvement.