Engineering a robust wnt-related diagnostic model for lung adenocarcinoma via machine learning algorithms and multi-cohort validation

Lung adenocarcinoma (LUAD) remains one of the leading causes of cancer-related deaths globally, necessitating more effective diagnostic and prognostic biomarkers. This study aims to identify and validate a robust diagnostic signature for LUAD using machine learning algorithms and multi-cohort analysis. We performed differential expression analysis on two LUAD datasets (GSE32863 and GSE116959) and intersected differentially expressed genes with the Wnt signaling pathway gene set to identify 10 hub genes. LASSO regression and support vector machine-recursive feature elimination (SVM-RFE) algorithms were employed to screen for candidate biomarkers. TMEM132A and CDH3 were identified as independent diagnostic indicators through multivariate logistic regression analysis. A diagnostic model based on these two genes achieved an area under the receiver operating characteristic curve (AUC) of 0.988 (95 % confidence interval: 0.968–1.000) in the training cohort and consistent performance across nine independent validation datasets (AUC range: 0.870–0.979). A nomogram was constructed to integrate the RiskScore with clinical variables, which significantly improved diagnostic accuracy compared to clinical variables alone. The model demonstrated exceptional early diagnostic potential for Stage I LUAD (AUC: 0.999). The expression of TMEM132A and CDH3 was validated at the protein and mRNA levels by immunohistochemistry and qRT-PCR, respectively. Immune infiltration analysis revealed that the diagnostic signature correlates with distinct tumor microenvironment profiles, including alterations in Th2 cells, mast cells, and dendritic cells. These findings establish a robust and practical diagnostic model with strong potential for clinical application in LUAD diagnosis and patient stratification.