Integrated bioinformatics and machine learning identify key hub genes in early-onset gastric cancer

Abstract

Early-onset gastric cancer (EOGC) represents a clinically and biologically distinct subtype of gastric cancer characterized by unique molecular features. However, its underlying molecular mechanisms remain unclear. This study aimed to identify EOGC-associated hub genes using an integrated bioinformatics and machine learning framework. Differentially expressed genes (DEGs) between EOGC and late-onset gastric cancer (LOGC) were analyzed using functional enrichment, weighted gene co-expression network analysis (WGCNA), least absolute shrinkage and selection operator (LASSO) regression, and random forest algorithms. Immune infiltration was characterized using CIBERSORT. Candidate hub genes were validated using an independent external cohort, clinical specimens, single-cell transcriptomic datasets, and in vitro functional assays. A five-gene logistic regression model was established and evaluated using ROC-AUC, calibration analysis, and bootstrap resampling. Transcriptomic analysis identified 227 DEGs, and WGCNA identified seven co-expression modules, with the green module showing the strongest association with EOGC. Machine learning identified five hub genes: CCR7, CD1D, PTGER2, PIK3CG, and EBI3. The combined model showed good discriminatory performance in the training, internal validation, and independent external validation cohorts (AUC = 0.909, 0.859, and 0.802, respectively). Bootstrap resampling yielded AUC estimates of 0.908, 0.852, and 0.789, respectively, supporting model stability. All five genes showed higher expression in EOGC and were validated at the mRNA and protein levels. Immune profiling and single-cell analysis revealed an immune-associated molecular phenotype and distinct expression patterns across immune cell subsets. Silencing the hub genes suppressed gastric cancer cell proliferation and migration in vitro . We identified five EOGC-associated hub genes and revealed an immune-associated molecular landscape. The five-gene model represents an EOGC-associated molecular signature distinguishing EOGC from LOGC rather than a stand-alone clinical diagnostic tool. These findings provide insights into the molecular and immune biology of EOGC and may inform future studies of molecular stratification and treatment-related outcomes.