From prediction to actionable mechanisms: Explainable multi‑omics AI for farm‑to‑fork postharvest preservation

Graphical overview of explainable artificial intelligence (XAI) for farm-to-fork postharvest preservation. Postharvest deterioration accumulates across orchard, packhouse, refrigerated transportation, warehouse, and distribution stages under fluctuating temperature, humidity, atmosphere, and mechanical stress. Multimodal data streams, including host omics, microbiome profiles, environmental sensing, RGB/hyperspectral/thermal imaging, spectroscopy, key genes, and logistics records, are integrated through a data lakehouse and analyzed by postharvest XAI models. Explainable modules, including SHapley Additive exPlanations (SHAP)/local attribution, graph neural network (GNN) explanation, pathway-constrained models, counterfactual reasoning, and stability/faithfulness auditing, convert black-box spoilage-risk prediction into interpretable biological mechanisms. These mechanisms guide actionable interventions such as antioxidant coating, elicitor spray, biocontrol consortia, packaging optimization, and gene-targeted strategies. Validation through storage trials, sensory evaluation, microbial testing, and sequencing closes the loop from prediction to explanation, intervention, and validated decision support.