An Epithelial-Mesenchymal Transition-Based Prognostic Model for Survival Prediction in Lung Adenocarcinoma: COL5A2 and ZEB2.

High heterogeneity and complex molecular mechanisms of lung adenocarcinoma (LUAD) lead to significant variability in patient prognosis. Epithelial-mesenchymal transition (EMT) is decisive to treatment response and tumor prognosis, which suggests that developing an EMT-related gene model might facilitate the prognostic management of LUAD. The RNA-seq, clinical information, and imaging data of LUAD patients were collected from public databases. EMT-related modules were clustered by performing WGCNA with the "WGCNA" package. Differentially expressed genes (DEGs) were identified using the "DESeq2" package. A RiskScore model was constructed via uni/multivariate Cox regression analyses and Lasso regression, implemented with the "survival" and "glmnet" packages. The prognostic performance was evaluated using the Receiver Operating Characteristic (ROC) curve with the "timeROC" package. The CIBERSORT algorithm was utilized to conduct immune infiltration analysis. Drug sensitivity analysis was performed with the "oncoPredict" R package, followed by employing GSEA_4.4.0 software to conduct GSEA. A nomogram model was developed utilizing the "rms" package. Finally, the expression and potential functions of the selected key genes were validated through cellular assays. A prognostic RiskScore model was developed based on two EMT-related genes (COL5A2 and ZEB2), which were identified through WGCNA and differential expression analysis. This model effectively classified LUAD patients into low- and high-risk categories, where those in the highrisk subgroup exhibited markedly reduced overall survival. Tumor immune microenvironment analysis revealed distinct infiltration patterns between the two risk groups. Notably, high-risk LUAD patients exhibited enrichment of oncogenic pathways, including EMT and E2F targets. Computational assessments further indicated that the high-risk group was associated with an increased possibility of immune evasion and lower sensitivity to certain chemotherapeutic agents. Additionally, a radiomics-based nomogram incorporating two CT features exhibited promising diagnostic performance. Finally, in vitro experiments demonstrated that silencing COL5A2 suppressed LUAD cell proliferation, migration, and invasion, confirming its oncogenic role in LUAD. The present work discovered and systematically validated the potential regulatory roles of COL5A2 and ZEB2 in LUAD progression. However, future prospective studies and experimental validations are still needed to confirm the clinical utility of the model. We developed an EMT-related prognostic model and a non-invasive assessment nomogram for LUAD patients, hoping to facilitate personalized therapy for LUAD.