Artificial intelligence in plant salt stress research: from predictive models to multi-omics integration

Salinity is a chronic environmental stressor causing irreversible damage to plants and resulting in significant economic losses. Early bioinformatics analyses on mono-omics data relying on predictive methods were highly effective in shedding light on the mechanisms of adaptation to salt stress. The incorporation of artificial intelligence has enabled analysis of multi-omics datasets combined with molecular, physiological, and morphological parameters relating to salt stress, and made it possible to perform high-throughput phenotyping using satellite snapshots and hyperspectral imaging to estimate soil salinization, predict salt stress in crops, and assess plant growth. Additionally, the arrival of transformers and the elaboration of large language models based on protein and nucleic acid sequences enabled identification of complex patterns underlying the 'language of life'. These generative models offer innovative hypotheses and experiments, particularly for understudied species or complex biological processes like salt stress tolerance. Protein language models also provided satisfactory results in identifying salt stress-related post-translational modifications. Predictive agro-climatic models are proving beneficial to the crop agriculture sector: they are expected to increase yields and reduce the time and costs involved in development or identification of commercially viable salt-tolerant cultivars. In conclusion, artificial intelligence is stimulating the discovery of novel facets of plant responses to salt stress, which is opening new frontiers in salinity research and contributing to previously unimaginable achievements.