Abstract
<title>Abstract</title> <p> Background Drought is a severe abiotic constraint that significantly reduces global peanut ( <italic>Arachis hypogaea</italic> L.) yield and quality. Although numerous transcriptomic studies have investigated drought stress, inconsistent results across individual, small-scale experiments often limit the generalization of findings. To identify conserved molecular signatures, this study integrated transcriptome meta-analysis, weighted gene co-expression network analysis (WGCNA), and machine learning across three independent RNA-seq datasets comprising 42 samples. Results The meta-analysis identified 2,499 meta-significant drought-responsive genes, from which a highly stringent subset of 276 direction-consistent core genes were derived. Using WGCNA on the broader gene set, the study identified a strongly drought-associated brown module enriched for key regulatory processes, including DNA-binding transcription factor activity, calcium ion binding, protein phosphorylation, and ubiquitin protein ligase activity. Machine learning analyses evaluated four classifiers and two feature-selection frameworks to test the predictive capacity of the 276 core genes. The results revealed that an ANOVA-based logistic regression model yielded the strongest predictive performance for distinguishing drought-stressed from well-watered samples, achieving an outer balanced accuracy of 0.9355. This model prioritized a set of recurrent candidate genes associated with signal transduction, post-translational regulation, osmotic adjustment, lipid biosynthesis, and structural remodeling. Conclusions Collectively, this computational framework identified a focused set of candidate drought-responsive genes from complex, multi-study transcriptomic data. These findings provide a useful foundation for future functional validation and for exploring candidate genes that may support genomics-assisted breeding of drought-resilient peanut cultivars. </p>