Development of a Diagnostic Model for Barrett's Esophagus and Esophageal Adenocarcinoma Based on Machine Learning and Immune Infiltration

Background Esophageal adenocarcinoma (EAC) is a highly lethal cancer, with Barrett's esophagus (BE) as its only known precursor. Early diagnosis is challenging, and the key biomarkers and mechanisms driving the BE-to-EAC progression are not fully understood. Methods We integrated transcriptomic data from BE (from the Gene Expression Omnibus (GEO)) and EAC (from The Cancer Genome Atlas (TCGA)) to identify shared differentially expressed genes (SDEGs). We then performed functional enrichment, protein-protein interaction, and immune infiltration analyses. Machine learning algorithms, including Least Absolute Shrinkage and Selection Operator (LASSO) and Random Forest, were applied to screen for diagnostic biomarkers. Results We identified 101 SDEGs. Subsequent analysis highlighted immune-related regulatory genes. Machine learning pinpointed MUC13 and PBLD as robust diagnostic biomarkers. The diagnostic model exhibited excellent performance, with area under the curve (AUC) values of 0.94 for BE and 0.98 for EAC, supported by nomograms and calibration curves (p > 0.9). Mechanistically, MUC13 promotes tumor progression via NF-κB, Wnt/β-catenin, MAPK/ERK, HIF-1α/VEGF, and epithelial-mesenchymal transition (EMT) pathways, driving inflammation-immune crosstalk and metastasis. In contrast, PBLD appears to suppress these processes. Conclusions MUC13 and PBLD are identified as potential biomarkers for the progression from BE to EAC. They play opposing roles in regulating key oncogenic pathways, immune response, and metastasis, offering significant potential for improving early diagnosis and developing targeted therapies.