AI-Driven Integration of Multi-Omics Data for Gene Function Discovery and Prediction of Complex Crop Phenotypes

Recent advances in high-throughput genotyping, phenotyping, and multi-omics technologies have generated large, heterogeneous datasets spanning genomics, transcriptomics, proteomics, metabolomics, epigenomics, and phenomics across plant species. Conventional statistical and machine-learning approaches often struggle to integrate these modalities, limiting mechanistic insight and predictive accuracy for complex genotype-environment-phenotype relationships. Artificial intelligence (AI), particularly deep learning and graph-based architectures, provides powerful tools for multi-omics data fusion, regulatory network reconstruction, and prediction of polygenic, environmentally modulated traits. This review synthesizes current applications of AI-driven multi-omics integration in plant science, with a focus on gene function annotation, cellular network inference, and complex crop phenotype prediction. We examine key methodological frameworks including autoencoders, variational generative models, multimodal transformers, graph neural networks, and self-supervised foundation models, highlighting representative case studies in Arabidopsis and major crops such as rice, maize, wheat, and tomato. We critically assess challenges related to data quality, batch effects, domain shift, metadata standardization, model interpretability, and reproducibility, and outline future directions encompassing plant-specific foundation models, pan-omics integration, digital-twin cropping systems, and community-driven benchmarking.