Construction and validation of immune prognosis model for lung adenocarcinoma based on machine learning.

INTRODUCTION

Lung adenocarcinoma is a leading subtype of lung cancer with high rates of recurrence and metastasis. Identifying novel prognostic biomarkers is essential for improving patient outcomes.

METHODS

Transcriptomic and clinicopathological data from TCGA (55 tumor samples and 38 normal samples) were used to construct a prognostic model, with 30 samples for internal validation. An external validation cohort (10 tumor-normal pairs) was obtained from the First Affiliated Hospital of Wenzhou Medical University. Differentially expressed genes and immune-related genes from the IMMPORT database were integrated using WGCNA. Three machine learning algorithms-Random Forest, LASSO, and SVM-RFE-were applied to identify key hub genes. A multivariate Cox regression model was built to predict survival. Model performance was assessed by time-dependent ROC and ANN models. Immune infiltration was analyzed using TIMER and ssGSEA, with consensus clustering performed to explore immune subtypes. Protein expression and biological functions of hub genes were validated using the HPA database and GSEA.

RESULTS

A total of 1,822 DEGs were identified, with 68 immune-related genes significantly associated with LUAD prognosis. Four hub genes-CBLC, GDF10, LTBP4, and FABP4-were selected to construct the prognostic model, which showed strong predictive performance in both ROC and ANN analyses. Immune profiling revealed elevated CD4⁺ T cells, macrophages, and dendritic cells in LUAD. Consensus clustering identified two immune subtypes with distinct prognoses and immune landscapes.

DISCUSSION

This study established a robust immune-related prognostic model for LUAD and identified key biomarkers associated with immune infiltration and survival. These findings offer valuable insights for personalized diagnosis and treatment strategies in LUAD.