Shapley values as metrics for studying genotype-by-environment interaction

The study of genotype-by-environment interaction is essential in plant breeding, as stability and adaptability determine the success of genotype recommendations. Traditionally, numerous statistical methods and stability indices have been employed, yet they often face limitations when dealing with complex and nonlinear relationships. In this study, we aimed to apply Shapley values (SHAP), derived from game theory and widely used in machine learning models, as an alternative approach for interpreting genotype-by-environment interaction. The mean SHAP value proved to be strongly associated with adaptability, reflecting the average performance of genotypes, while the variability of SHAP values emerged as a reliable indicator of stability. Moreover, the method allowed an integrated assessment of environmental contributions, identifying favorable and unfavorable sites. Although it showed moderate agreement with traditional metrics, SHAP provided complementary insights, enhancing the understanding of genotype-by-environment interaction. Therefore, these metrics represent a promising tool to support decision-making in plant breeding programs, particularly for identifying superior genotypes across diverse environments.