Diagnostic Biomarkers of Metabolic Reprogramming-Related Genes for Gastroesophageal Reflux Disease: Machine Learning Modeling, Molecular Mechanisms, Drug Prediction, and Experimental Validation

Background: Metabolic reprogramming plays a critical role in the pathogenesis of gastroesophageal reflux disease (GERD). This study aimed to identify metabolic reprogramming-related diagnostic biomarkers, construct a machine learning-based diagnostic model, and explore the underlying mechanisms and potential therapeutic agents. Methods: GERD transcriptomes from the Gene Expression Omnibus (GEO) were corrected for batch effects, and candidate genes were derived by overlapping differentially expressed genes, weighted gene co-expression network analysis (WGCNA) modules, and metabolic reprogramming gene sets. Models built with 107 machine-learning algorithms were ranked by the area under the receiver operating characteristic(ROC) curve (AUC), and the best-performing model was verified by an artificial neural network (ANN); a nomogram was then constructed from the core genes. Mechanisms were explored through protein–protein interaction (PPI) network analysis, functional enrichment, and CIBERSORT immune infiltration analysis. Potential therapeutic agents were predicted by AI-based drug screening and molecular docking. Finally, a rat model of GERD was established, and in vivo validation was performed using real-time PCR, western blot, and immunohistochemistry. Results: The glmBoost+GBM ensemble model achieved the best performance using the seven screened candidate genes. All seven genes yielded AUC values > 0.7, and the top five genes ranked by AUC (CKB, SH3GL2, JUP, LCN2, and CPA2; AUC values > 0.80) were selected as core genes. ANN validation supported the model’s robustness. The nomogram showed good calibration and net clinical benefit. Enrichment analysis revealed pathways associated with cell adhesion, amino acid metabolism, and IL-17 signalling. Immune infiltration analysis indicated immune imbalance, and core genes were closely correlated with immune cell subsets. Three compounds with high binding affinity (alvimopan, baricitinib, and isoevodiamine) were identified, and the differential expression of all five genes in GERD was confirmed by animal experiments. Conclusions: CKB, SH3GL2, JUP, LCN2, and CPA2 are core diagnostic biomarkers of metabolic reprogramming in GERD. The diagnostic model and nomogram exhibit excellent discriminative performance.These genes may participate in GERD pathogenesis through metabolic–immune crosstalk, providing novel targets and a theoretical basis for early diagnosis, risk stratification, and targeted therapy.