A Combined Machine Learning and Weighted Gene Co-expression Network Analysis Strategy for Lung Adenocarcinoma Biomarker Discovery

Lung adenocarcinoma (LUAD), the most common subtype of non-small cell lung cancer (NSCLC), remains a leading cause of cancer-related deaths worldwide, largely due to difficulties in early detection and variable treatment responses. Identifying molecular biomarkers that predict disease progression and therapeutic outcomes is essential for advancing precision oncology in LUAD. In this study, we applied an integrative approach combining machine learning and bioinformatics to identify survival-associated genes and develop a robust prognostic risk model. Multiple machine learning algorithms – including Cox Proportional Hazards (CoxPH), TabNet, TabNetCoxPHAdd, and TabNetCoxPHMult – were used for gene selection, while Weighted Gene Co-expression Network Analysis (WGCNA) identified hub genes linked to patient survival. Genes detected by both methods were refined through CoxPH analysis, yielding 14 key prognostic genes: PDCD5, PSMB1, AHCY, CYC1, GOLT1B, COPS9, ATP5F1C, HPRT1, COX5B, PDCD10, SDHD, NDUFB5, RP53, and RACK1. Functional interactions among these genes were explored through a protein–protein interaction network constructed using the STRING database. The resulting risk prediction model showed high predictive accuracy and reproducibility when validated on independent Gene Expression Omnibus (GEO) datasets. These results identify potential prognostic biomarkers for LUAD and provide a validated framework for personalized treatment strategies, supporting the advancement of precision oncology in this disease.