Predicting enviromically adapted varieties with big data.

Area of Science:

• Agricultural Science
• Genetics
• Plant Breeding

Background:

• Traditional breeding programs focus on average genotype performance, potentially missing environment-specific adaptations.
• Selecting genotypes for specific environments is crucial for optimizing crop yields.

Purpose of the Study:

• To develop a genomic prediction framework for selecting high-yielding winter wheat genotypes tailored to individual environments.
• To improve the prediction of genotype-specific performance by accounting for genotype-by-environment interactions.

Main Methods:

• Compiled extensive winter wheat grain yield data from 13,285 genotypes across 31 Central European sites (2010-2022).
• Utilized convolutional neural networks (CNNs) and traditional genomic best linear unbiased prediction (GBLUP) for predicting genotype performance.
• Incorporated environmental data to model genotype-by-environment (G×E) interactions using machine learning.

Main Results:

• CNNs demonstrated competitive to superior performance compared to GBLUP for predicting average genotype performance as training data size increased.
• A 23% improvement in predicting environment-specific hybrid performance was observed using GBLUP models with G×E interactions.
• Identified key environmental variables driving G×E interactions and genotype clustering across Central European study sites.

Conclusions:

• Big data, machine learning, and deep learning offer novel approaches to overcome genetic bottlenecks in crop breeding.
• These advanced methods facilitate faster development and delivery of improved crop varieties to farmers.