Closing the loop: AI-driven integration of multi-omics and phenomics for systematic resilient crop engineering

Sustaining global food security amidst accelerating climate change necessitates a paradigm shift from reductionist breeding to systematic, AI-driven crop engineering. This review provides a critical synthesis of the edge-to-cloud closed-loop framework that unifies multi-omics, real-time phenomics, and synthetic biology. We first delineate the methodological evolution from linear statistical models to deep representation learning, highlighting how transformer architectures and graph neural networks decode the non-linear dynamics of genotype-environment-management interactions. Specifically, we examine the deployment of multimodal fusion algorithms in deciphering complex stress mechanisms, ranging from molecular regulatory networks to the holobiont-level responses under drought, salinity, and combined stressors. Crucially, the review articulates the transition from passive prediction to proactive design, where generative AI (GAI) and foundation models accelerate the Design-Build-Test-Learn cycle by designing de novo stress-resilient proteins and gene circuits. Finally, we propose a roadmap to bridge the in silico-to-in situ gap, advocating for pan-genomic foundation models, physics-informed causal AI to ensure biological interpretability, crop digital twins that continuously synchronize simulation with field reality, and the establishment of autonomous self-driving laboratories. This systemic integration aims to transform crop improvement into a predictive engineering discipline capable of delivering resilient cultivars for volatile future climates.