Machine learning–guided integrative analysis identifies phycobilisome hub genes underlying abiotic stress responses in Synechocystis sp. PCC 6803

Identifying conserved genetic regulators that underpin the general abiotic stress response in cyanobacteria remains a major challenge due to stress-specific transcriptional variability and platform-dependent biases. Here, we present an integrative systems-biology framework that combines transcriptomic meta-analysis, weighted gene co-expression network analysis (WGCNA), and machine learning to dissect stress-responsive regulatory hubs in Synechocystis sp. PCC 6803 . Analysis of 116 microarray samples spanning diverse abiotic stresses identified 62 robust differentially expressed genes, including strong induction of sll0218 (log₂FC = 2.88) and pronounced repression of sll1471 (log₂FC = −2.42). WGCNA resolved 13 co-expression modules, of which the blue and red modules were strongly negatively correlated with stress (r = −0.71 and −0.67, respectively), whereas the yellow module showed a positive association (r = 0.67). Random Forest and XGBoost classifiers achieved high predictive performance (accuracy > 96%, AUC > 0.99) and consistently prioritized a small subset of highly informative genes. Cross-integration of all analytical layers converged on three phycobilisome-related hub genes— slr1986 ( apcB ), sll1577 ( cpcB ), and slr2067 ( apcA ). Experimental validation by RT-qPCR revealed strong downregulation of these genes under salt stress, with log₂FC values of approximately −4.3 ( apcB ), −2.9 ( cpcB ), and −3.5 ( apcA ). Notably, an independent RNA-seq meta-analysis of high-light stress datasets further confirmed their significant repression (log₂FC range: −2.00 to −2.81), demonstrating platform-independent robustness. Collectively, this study provides the first multi-layer, cross-platform evidence that coordinated repression of phycobilisome components constitutes a conserved and central pillar of the general stress response in Synechocystis , establishing these genes as robust and transferable biomarkers of cellular stress. • An integrative pipeline combines meta-analysis, WGCNA, and machine learning. • Three phycobilisome genes ( apcA , apcB , cpcB ) are core stress biomarkers. • Phycobilisome dismantling is a primary pillar of the general stress response. • Machine learning models classify stress conditions with >96% accuracy.