Using machine learning to combine genetic and environmental data for maize grain yield predictions across...

Area of Science:

• Agricultural Science
• Genetics
• Machine Learning

Background:

• Genomic prediction models are crucial for crop breeding.
• Integrating environmental data can improve prediction accuracy.
• Understanding genotype-by-environment interactions (GEI) is vital for optimizing crop performance.

Purpose of the Study:

• To explore novel machine learning approaches for combining non-genetic (environmental) information into genomic prediction models.
• To evaluate the efficacy of feature-engineered environmental data in enhancing prediction accuracy for maize grain yield.
• To compare additive (G+E) and multiplicative (GEI) modeling strategies for integrating genetic and environmental factors.

Main Methods:

• Utilized multi-environment trial data from the Genomes To Fields initiative.
• Developed and compared machine learning models using genetic data, environmental data, or a combination.
• Implemented additive (G+E) and multiplicative (GEI) approaches for data integration.
• Employed feature engineering to process high-dimensional environmental data (climate, soil).

Main Results:

• Machine learning models incorporating environmental data increased mean prediction accuracy by up to 7% compared to a standard model.
• The additive G+E model demonstrated superior or comparable prediction accuracy to the GEI model.
• The G+E model offered advantages in computational efficiency (memory and time) and flexibility.
• Feature engineering proved effective for envirotyping and generating valuable data for genomic prediction.

Conclusions:

• Feature-engineered environmental data integrated into machine learning models efficiently capture GEI effects indirectly.
• The G+E approach is a flexible and efficient strategy for merging genotypic and environmental data in breeding programs.
• Machine learning, particularly with feature engineering, offers a powerful framework for enhancing genomic prediction by leveraging diverse data sources.