Machine learning and functional validation identify OsRAV11/12 as negative regulators of drought tolerance and early...

Rice (Oryza sativa L.) is a staple food for half of the global population, with drought stress posing a significant threat to production. This study employed machine learning (ML) approaches on meta-transcriptomic data to identify key regulatory genes involved in drought stress response. We trained Random Forest (RF), XGBoost, and Feedforward Neural Network (FNN) models on rice transcriptomic data, with XGBoost demonstrating superior performance (90 % accuracy, 0.97 AUC). ML predicted both known drought tolerance-related genes (e.g., OsCCA1, OsPYL1, and OsNAC22, etc.) and previously uncharacterized candidates (e.g., OsFBX187, LOC_Os01g70010, LOC_Os02g24750, and LOC_Os03g23950, etc.), with OsLTRPK1 and OsCYC emerging as the most influential predictors of drought response. Notably, OsRAV11 exhibited a strong inverse association with drought tolerance, suggesting its role as a negative regulator. Functional validation through CRISPR/Cas9-mediated knockout (KO) and overexpression (OX) confirmed that OsRAV11/12 negatively regulates drought tolerance. In addition, KO lines demonstrated enhanced drought resilience with 80-85 % survival rates compared to wild-type (WT; 57.5 %), while OX lines showed increased susceptibility. Physiological analysis revealed that KO lines maintained lower stomatal conductance, accumulated higher abscisic acid (ABA) levels, and exhibited reduced lipid peroxidation under drought conditions. Transcriptome analysis revealed upregulation of stress response and protein folding processes in KO lines, whereas OX lines exhibited inappropriate upregulation of growth-related processes and downregulation of membrane integrity genes. KEGG pathway analysis further supported these observations, with KO lines showing enrichment in metabolic pathways, secondary metabolite biosynthesis, and plant hormone signal transduction. Our findings establish OsRAV11/12 as negative regulators of drought tolerance in rice and demonstrate the power of ML in accelerating gene discovery for complex traits.