Jiaying Wu, Zhen Zheng, Jiaming Wu
Barrett's esophagus (BE) is a precursor of esophageal adenocarcinoma, but candidate expression markers and preventive hypotheses require evaluation across patients, cohorts, and evidence levels. We analyzed two GPL96 discovery cohorts comprising 15 matched normal-BE pairs, used pair-aware differential modeling and resampling, and rebuilt four external BE-tissue assessments from official GEO files. The primary analysis yielded 1,085 differentially expressed genes; LASSO selected 13 features, linear SVM-RFE selected 6, and pair-bootstrap Random Forest retained 20. DPP4 and CCRL2 were shared by the three primary feature sets, but not after phenotype-preserving ComBat sensitivity analysis. External Affymetrix AUCs were 1.000, 0.855, and 0.976 for DPP4 and 1.000, 0.858, and 0.963 for CCRL2. In the Illumina GSE28302 cohort, fixed-direction DPP4 AUC was 0.444 and CCRL2 lacked a uniquely mapped probe. Seventy-nine DEGs intersected the verified 710-gene disease union; STRING/CTD analysis ranked Niuxi, Mahuang, and Mahuanggen first through third. Reference-controlled docking advanced kaempferol-PPARG as the primary structural hypothesis and taxifolin-NQO1 as a secondary extension. Twelve 100 ns trajectories retained pocket-region contacts, but query compounds showed greater pose rearrangement than the reference ligands. These findings prioritize candidates for experimental validation and do not establish a diagnostic test, direct binding, expression reversal, or therapeutic efficacy.