A novel matrix multiplication framework for modeling genotype-by-environment interaction in genomic prediction.

Matrix-product-based kernels improve genomic prediction of genotype-by-environment interactions, with up to 13.2% higher predictive accuracy than Hadamard kernels in wheat and rice datasets.

• Why it matters: Accurately modeling G × E interaction is crucial for plant breeding across diverse environments, but current methods like Hadamard products may not fully capture the complex structured relationships observed in multi-environment trials.
• What they did: Researchers extended a matrix-product framework to explicitly model G × E interaction by deriving symmetric kernels from genotype and environment covariance matrices, evaluated on eleven multi-environment datasets in wheat and rice.
• The result: Models with matrix-product-derived kernels showed higher predictive performance than conventional Hadamard models, suggesting these kernels carry complementary information, though their interpretation requires consideration of structural limitations.