Identification of key genes as diagnostic biomarkers for lung adenocarcinoma using bioinformatics and machine learning

Lung adenocarcinoma (LUAD) is the most prevalent histological subtype of lung cancer; however, stable and reliable molecular diagnostic markers remain elusive. This study aimed to identify key genes with diagnostic utility and clinical relevance using bioinformatics and machine learning approaches. Transcriptomic data from GEO datasets (GSE10072, GSE116959, GSE32863) were integrated with batch-effect correction. Differentially expressed genes (DEGs) were identified and analyzed by weighted gene co-expression network analysis (WGCNA); their overlap defined candidate genes for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment. Summary data–based Mendelian randomization (SMR) using FinnGen lung adenocarcinoma data highlighted eight overlapping genes. The integrated GEO cohort was used for training, with GSE43458 and GSE75037 for external validation, to build machine-learning models and derive a five-gene diagnostic signature. Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts (CIBERSORT) assessed immune infiltration and gene–immune associations, and prognostic/clinical relevance was validated in The Cancer Genome Atlas (TCGA-LUAD). Five key genes with excellent diagnostic performance (AUC > 0.8 in both the training and validation cohorts) were identified. These genes were closely associated with immune cell infiltration and immune-related features, and several showed significant correlations with overall survival and clinicopathological characteristics. Colocalization analysis indicated that ACVRL1 had a PP.H4 of 0.53 and a PP.H4/(PP.H3 + PP.H4) ratio of 0.97, suggesting a potential causal association between ACVRL1 and lung adenocarcinoma risk, with an inverse correlation. By integrating multi-cohort transcriptomic data, SMR analysis, and machine learning approaches, this study identified five key genes with diagnostic potential for lung adenocarcinoma. Their immunological characteristics and clinical relevance were systematically evaluated. Colocalization analysis further suggested that ACVRL1 may be inversely and potentially causally associated with lung adenocarcinoma risk, providing new insights into lung adenocarcinoma pathogenesis and facilitating the identification of candidate molecular biomarkers.