Machine learning identifies palmitoylation related gene signatures for biomarker discovery and drug repurposing in lung cancer

Lung cancer remains a leading cause of cancer-related death. Palmitoylation, a critical post-translational modification, regulates key oncogenic pathways, yet its mechanistic role in lung carcinogenesis is poorly understood. This study aims to identify robust palmitoylation-related targets with biological and clinical significance and to explore their therapeutic potential, focusing on their roles in cell cycle regulation and immune modulation. Three lung cancer microarray datasets (GSE10072, GSE19188, and GSE30219) were used as the training cohort ( n = 570; 442 tumor and 128 normal samples), and two independent cohorts (GSE74706, n = 36; GSE68465, n = 443) were used for validation. Batch effects were corrected with ComBat incorporating a biological design matrix, and differentially expressed genes (DEGs) were identified with the limma package (|log2FC| > 1, adjusted p < 0.05). A curated set of 350 palmitoylation-related genes (240 from GeneCards, 110 from the literature) was intersected with the DEGs to define palmitoylation-related DEGs(PR-DEGs). To minimize algorithm-specific bias, three complementary machine-learning methods with different underlying assumptions were applied-LASSO regression (linear, L1-sparse), support vector machine recursive feature elimination (SVM-RFE), and random forest (nonlinear ensemble)-and the consensus intersection of the three was retained. The clinical relevance of the selected genes was assessed by ROC analysis in the training cohort and in the GEO and TCGA validation cohorts. Immune infiltration was profiled by ssGSEA and xCell.Candidate therapeutic compounds were predicted with the DSigDB database, and molecular docking was performed using AutoDock Vina (CB-Dock2). Three algorithms identified PLA2G1B (downregulated) and DSG2 (upregulated) as robust feature genes. Pathway enrichment indicated their involvement in cell cycle regulation, while immune profiling revealed a significant positive correlation between DSG2 and CD8 + T cell infiltration ( r = 0.63, p < 0.01). The combined model showed excellent diagnostic performance (training: AUC = 0.907;GSE74706:AUC = 0.87;GSE68465:AUC = 0.86;TCGA-LUAD:0.943; nnTCGA-LUSC: 0.998). High-risk (low PLA2G1B/high DSG2) patients had poorer overall survival (HR = 2.34, 95% CI: 1.52–3.61, p < 0.001). Drug prediction and molecular docking identified Montelukast sodium as a promising candidate for repurposing, with high predicted binding affinity to PLA2G1B (-6.9 kcal/mol). This study computationally nominates PLA2G1B and DSG2 as candidate palmitoylation-related biomarkers. Their potential roles in cell cycle and immune pathways, as well as the predicted binding of montelukast to PLA2G1B, provide testable hypotheses that require rigorous experimental validation.