Integrated transcriptomic analysis and machine learning identify immunogenic cell death genes as prognostic markers and therapeutic targets in non-small cell lung cancer.

Background: Immunogenic cell death (ICD) is a regulated cell death that activates antitumor immunity, yet its prognostic role in non-small cell lung cancer (NSCLC) remains unclear. This study aimed to develop and validate a robust ICD-related gene (ICDRG) signature for predicting survival and characterizing the tumor immune microenvironment in NSCLC. Methods: In this retrospective prognostic model development and validation study, transcriptomic and clinical data from 598 NSCLC patients in The Cancer Genome Atlas (TCGA) cohort were used for model training. External validation was performed using three independent Gene Expression Omnibus (GEO) cohorts [GSE11969, GSE68465 and GSE81089, total n=441 from GSE11969/GSE68465, with GSE81089 providing an additional RNA sequencing (RNA-seq)-based validation set]. Based on 34 literature-curated ICD genes, we identified prognostic candidates through single-cell sequencing analysis and weighted gene co-expression network analysis (WGCNA). An 8-gene prognostic signature based on the ICD-related risk score (ICDRS) was constructed using least absolute shrinkage and selection operator (LASSO)-Cox regression and validated with a machine learning (ML) ensemble framework (10 algorithms). Model performance was assessed using Kaplan-Meier analysis, time-dependent receiver operating characteristic (ROC) curves, and multivariate Cox regression. Associations between the ICDRS and immune infiltration or drug sensitivity were further evaluated. Results: The ICDRS model stratified patients into high- and low-risk groups with significantly different overall survival (OS) in both the training and validation cohorts (all log-rank P<0.05). The model demonstrated robust predictive accuracy for 1-, 3-, and 5-year survival [area under the curve (AUC) >0.70]. Multivariate analysis confirmed the ICDRS as an independent prognostic factor [hazard ratio (HR) >2.0, P<0.001]. Notably, consistent prognostic performance was observed across microarray-based (GSE11969/GSE68465) and RNA-seq-based (GSE81089) platforms, underscoring the signature's robustness to technical variations. Furthermore, the high-risk group was characterized by an immunosuppressive microenvironment and higher predicted resistance to common chemotherapeutic agents. Conclusions: We developed and validated an 8-gene ICD-related signature that serves as an independent prognostic biomarker for NSCLC. This model provides insights into the immunogenic landscape of tumors and offers a potential tool for personalizing immunotherapy strategies.