Deep learning genomic selection for disease-resistant crop breeding

Overview

This solution applies deep learning models to genomic selection in crop breeding, enabling more accurate prediction of disease resistance and yield in wheat, oat, and chickpea. By capturing both major and minor gene effects through machine learning, the approach supports the development of durable disease-resistant crop varieties while maintaining genetic diversity. It is designed to integrate into broader integrated pest management strategies, offering plant breeders and agricultural companies a data-driven tool to accelerate variety development and improve selection decisions.

Technical specifications

• Deep learning architecture developed in collaboration with the Australian Institute for Machine Learning, specifically designed to handle large-scale genomic and phenotypic datasets
• Hybrid gene modeling that combines major gene effects with minor gene contributions to produce more accurate predictions of complex traits such as disease resistance and grain yield
• Validated on a large-scale wheat dataset comprising approximately 300,000 breeding lines and approximately 25,000 DNA markers provided by Australian Grain Technologies (AGT), the largest plant breeding company in Australia
• Benchmarked against standard mixed-linear models, showing significantly decreased prediction error for grain yield
• Adaptation pathway for smaller datasets in oat (~500 breeding lines) and chickpea (~1,000 breeding lines), extending applicability beyond large-scale cereal breeding programs
• Genetic diversity preservation built into the selection framework to support breeding strategies that maintain broad genetic variation in crop populations

Technology readiness level

The deep learning model is at a prototype stage. It has been tested with a large wheat dataset and demonstrated significantly improved prediction accuracy over conventional mixed-linear and Bayesian models. The team is now extending the model to smaller datasets for oat and chickpea, where existing grain yield, disease resistance, and DNA marker data are already available. Future validation will benchmark the new deep learning model against existing statistical approaches and explore the relationship between genomic selection and genetic diversity maintenance. The technology is ready for collaborative refinement and pilot deployment with breeding partners.