Harnessing meta-analysis and artificial intelligence to reveal conserved regulatory biosignatures of abiotic stress in soybean

BACKGROUND: Soybeans are widely cultivated worldwide as an important source of edible vegetable oil and protein. Due to climate change, it is repeatedly exposed to various abiotic stressors in its natural habitat. Abiotic stresses such as heat, drought, and salinity severely restrict soybean productivity, yet the conserved molecular mechanisms underlying multi-stress tolerance remain poorly understood. The integrated application of machine learning and co-expression network analysis for robust biosignature and hub gene discovery remains limited. Therefore, this study aimed to identify conserved stress-responsive biosignatures and explore their evolutionary and regulatory significance. RESULTS: Here, we explored the transcriptional regulation of soybean under multiple abiotic stress conditions, including heat, drought, and salt. A total of 14,503 genes are differentially expressed across three stress conditions, with 466 genes common to all three. Gene Ontology and KEGG pathway analyses indicated that the meta-DEGs primarily participate in oxidative stress, hormone signaling, and metabolic pathways. Segmental duplication is the key driving force of stress response gene expansion, and most of these expansions occurred through the recent whole-genome duplication (WGD) in soybean. The 12 abiotic stress-responsive biosignatures were identified using a wedge co-expression network and machine learning (ML)- based hub genes. A deep neural network (DNN) model was constructed to validate stress biosignatures, achieving 97.39% and 76.47% prediction accuracies on the test and external validation sets, respectively. CONCLUSIONS: Our findings revealed conserved stress-responsive genes, key regulatory hubs, and oxidative stress as a central molecular feature governing multi-stress adaptation. The integration of artificial intelligence enabled accurate validation of biosignatures, offering valuable insights into functional genomics and genomic-assisted breeding strategies. This study offers a strong foundation for AI applications in plant breeding and supplies valuable resources for soybean genetic improvement.