Predictive Breeding for Crop Genetic Improvement

Genomic prediction has revolutionized breeding by accelerating genetic gain, yet its accuracy can be significantly enhanced by effectively integrating environmental data, particularly to account for complex genotype-by-environment (G×E) interactions. This chapter provides a comprehensive overview of methodologies for leveraging environmental information in genomic prediction models. It explores various approaches, categorized into linear, non-linear and hybrid models. Linear models, including the direct inclusion of environmental covariates, environmental clustering and G×E interaction models (such as reaction norms and multi-trait models utilizing environmental relationship matrices), offer interpretable frameworks for capturing G×E. Non-linear approaches, including kernel-based methods and deep learning models, are highlighted for their ability to capture complex non-additive genetic effects and complex G×E patterns, with examples demonstrating the fusion of diverse data types. The chapter also explores hybrid models, specifically the integration of deep learning with linear mixed models, and ensemble methods, which combine the strengths of different modeling paradigms for improved accuracy and robustness. Finally, future directions emphasizing the integration of diverse omics and phenomics data, advanced AI approaches and the development of user-friendly tools to bridge the gap between research and practical application in breeding are discussed. Such strategies are crucial for developing climate-resilient crops and advancing sustainable agriculture.