AI-enabled multi-omics integration for predictive food safety

The growing complexity of global food systems and the burden of foodborne diseases highlight limitations in conventional food safety approaches based on retrospective testing and isolated analytical measurements. High-throughput omics technologies, including genomics, transcriptomics, proteomics, metabolomics, and microbiomics, enable comprehensive molecular characterization of food systems, but their scale and heterogeneity challenge conventional analytical approaches. Artificial Intelligence (AI) can integrate these heterogeneous datasets to support predictive food-safety, quality-assurance, and risk-assessment applications. This review provides a critical overview of AI-enabled multi-omics integration for food safety, covering major omics platforms, multimodal integration, machine learning (ML), deep learning (DL), graph neural networks (GNNs), transformer-based approaches, and explainable AI (XAI). Applications include pathogen detection and characterization, antimicrobial-resistance (AMR) surveillance, shelf-life and spoilage prediction, food authenticity assessment, and risk assessment. Reported results include 0.94 accuracy ( κ = 0.82 ) for genomic prediction of Salmonella disease-endpoint phenotypes, AUROC of 0.904 for Salmonella stress-response prediction, and R 2 > 0.96 for neural-network-based shelf-life prediction. However, performance remains dependent on data quality, model design, and validation strategy. Key challenges include data standardization, missing modalities, batch effects, model interpretability, regulatory acceptance, computational scalability, and limited availability of well-characterized multi-omics food-safety datasets. Emerging directions such as biological foundation models, digital twins, and cyber-physical systems are discussed as longer-term opportunities, while autonomous food-safety management remains an emerging research direction.