Prediction from genomics to fields: plant model-centred integration for precision breeding

Genotype-to-phenotype (G2P) prediction is the central mission of modern plant breeding. Genomic data accumulation has outpaced the ability of current frameworks to predict field performance across variable environments. Statistical genomic models deliver computationally efficient breeding values but lose accuracy when applied outside training environments. Artificial intelligence extracts high-dimensional patterns at very high throughput but extrapolates unreliably under novel stress conditions. Conventional crop models encode physiological rules across environments but treat the canopy as a homogeneous layer and cannot resolve architectural variation. Functional-structural plant models simulate three-dimensional organ-level morphogenesis at the specific biological scale where genetic variation in plant architecture shapes resource capture and yield. The four paradigms address distinct layers of the G2P problem and are therefore complementary rather than substitutable. Here we propose a plant-model-centred integration in which functional-structural plant models are the bidirectional hub linking upstream genomic parameter estimation, lateral multi-omics constraints, AI-based pattern recognition, and downstream canopy and field upscaling. We outline the structural rationale, review graded empirical evidence, and identify the technical conditions and benchmark trial needed to advance this framework towards operational precision breeding.