Deep Learning for Predicting Complex Traits in Spring Wheat Breeding Program.

Area of Science:

• Plant breeding
• Genomics
• Machine learning

Background:

• Genomic selection (GS) is crucial for plant breeding, but improved prediction accuracy for complex traits is needed.
• Analytical methods from other disciplines, such as deep learning (DL), offer potential advancements for GS.
• Deep learning utilizes artificial neural networks for sophisticated model training.

Purpose of the Study:

• To evaluate the potential of deep learning (DL) models in the Washington State University spring wheat breeding program.
• To compare the predictive performance of two DL algorithms (MLP and CNN) against a standard GS model (rrBLUP).

Main Methods:

• Employed multilayer perceptron (MLP) and convolutional neural network (CNN) deep learning algorithms.
• Utilized a dataset of 650 spring wheat recombinant inbred lines (RILs) from a NAM population (2014-2016).
• Predicted five quantitative traits using cross-validations, independent validations, and SNP markers, optimizing hyperparameters for DL models.

Main Results:

• Deep learning models achieved 0-5% higher prediction accuracy than the rrBLUP model across all five traits and validation strategies.
• The MLP deep learning model outperformed the CNN model, showing 5% higher accuracy for grain yield and protein content.
• Both MLP and CNN demonstrated superior prediction accuracy compared to rrBLUP for all traits studied.

Conclusions:

• Deep learning approaches provide enhanced prediction accuracy for complex traits in plant breeding.
• DL models, particularly MLP, should be integrated into plant breeder toolkits for large-scale breeding programs.
• These findings highlight the transformative potential of AI in accelerating crop improvement.