The Role of Immunogenic Cell Death in the Prognosis and Development of Treatment Strategies for Non-Small Cell Lung Cancer: A Multiomics and Machine Learning Approach

Transl Lung Cancer Res 2025 AI 7 Explanations View Original
Original Paper (PDF)

Unable to display PDF. Download it here or view on PMC.

Plain-English Explanations
Pages 1-2
Building a Smarter Prognostic Tool for NSCLC Using Immunogenic Cell Death

The Challenge Non-small cell lung cancer (NSCLC) is the most common form of lung cancer, and its highly variable genomic landscape makes predicting patient outcomes and selecting the right treatment extremely difficult.

What Is Immunogenic Cell Death? Immunogenic cell death (ICD) is a special form of cell death that triggers an immune response against the tumor. When cells die through ICD, they release signals that activate dendritic cells and ultimately stimulate tumor-specific immunity - making the tumor visible to the immune system.

The Gap While ICD has been linked to better cancer outcomes in other settings, its specific role in NSCLC prognosis and treatment response had not been systematically studied, leaving clinicians without a reliable ICD-based tool.

The Solution Researchers developed an ICD-related signature (ICDRS) by combining multiomics data from TCGA and GEO databases with 101 combinations of 10 different machine learning algorithms to find the most robust prognostic gene set.

TL;DR: This study used machine learning and multiomics data to build a gene signature based on immunogenic cell death that can predict NSCLC patient survival and treatment response.
Pages 2-4
101 Machine Learning Combinations to Find the Best Prognostic Genes

Data Sources Clinical and transcriptomic data came from The Cancer Genome Atlas (TCGA), and single-cell RNA sequencing data came from the Gene Expression Omnibus (GEO). Multiple external cohorts (GSE13213, GSE31210, GSE72094) were used for validation.

Gene Screening Strategy Weighted gene coexpression network analysis (WGCNA) was used to identify gene modules strongly associated with ICD. These were intersected with lung cancer prognostic genes and differentially expressed genes from TCGA, yielding 29 candidate ICD-related genes.

Machine Learning Ensemble Ten algorithms - including random survival forest, LASSO, ridge regression, CoxBoost, and gradient boosted regression - were combined in 101 configurations and cross-validated to find the most stable and accurate ICDRS.

Validation The resulting signature was validated at the genomic and single-cell transcriptomic levels. Gene expression was further confirmed with qRT-PCR and immunohistochemistry (IHC) on NSCLC tissue samples from real patients.

TL;DR: The team used 101 machine learning algorithm combinations alongside WGCNA gene network analysis to systematically identify and validate the most reliable ICD-related prognostic genes in NSCLC.
Pages 5-7
Five Key Genes Define High-Risk and Low-Risk NSCLC Patients

The Five Genes The final ICDRS was built from five genes: MMP14, ALDH2, FBP1, HLA-DRA, and KCTD12. Patients were split into high-risk and low-risk groups based on their expression levels.

Survival Differences High-risk patients had significantly worse prognosis compared to low-risk patients across all validation datasets, with Kaplan-Meier curves showing clear and statistically significant separation between the groups.

Immune Landscape Low-risk patients showed greater infiltration of anti-cancer immune cells including T cells, B cells, and NK cells, suggesting a more immunologically active tumor microenvironment that is more responsive to immunotherapy.

Drug Sensitivity Low-risk patients also showed better predicted response to both chemotherapy and targeted therapy (P less than 0.05), making the ICDRS a dual-purpose tool for prognosis and treatment selection.

TL;DR: A five-gene ICD signature stratifies NSCLC patients into risk groups with distinct survival outcomes and immune profiles, predicting responses to both immunotherapy and chemotherapy.
Pages 7-9
How the Tumor Microenvironment Differs Between ICDRS Risk Groups

Immune Cell Infiltration CIBERSORT and XCell analyses showed that low-risk patients had significantly higher levels of tumor-infiltrating immune cells including cytotoxic T cells and natural killer cells, which are key for anti-tumor defense.

Immunosuppression in High-Risk Patients High-risk patients showed enrichment of immunosuppressive cell types and lower ESTIMATE immune scores, reflecting a colder tumor immune environment that tends to resist immune checkpoint inhibitors.

TIDE Scores The tumor immune dysfunction and exclusion (TIDE) analysis predicted greater immune dysfunction in the high-risk group, consistent with lower expected benefit from immunotherapy.

Immune Checkpoint Expression Differential expression of immune checkpoint genes between the two groups was analyzed using the Wilcoxon test, providing additional mechanistic insight into why risk groups respond differently to ICI therapy.

TL;DR: Low-risk NSCLC patients identified by the ICDRS have hotter, more immune-active tumors, while high-risk patients show immunosuppressive microenvironments that resist checkpoint inhibitor therapy.
Pages 9-10
Mutational Landscape and Tumor Mutational Burden Across ICDRS Groups

Mutational Profiles The maftools R package was used to generate waterfall plots showing the distinct mutational landscapes of high-risk and low-risk NSCLC patients, revealing differences in the frequency and types of somatic mutations.

Tumor Mutational Burden TMB was calculated and survival analysis showed that the ICDRS effectively predicted TMB levels. Low-risk patients tended to have higher TMB, which is associated with greater neoantigen load and immunotherapy responsiveness.

Copy Number Variation Comprehensive CNV analysis highlighted differential amplification and deletion patterns of key ICDRS genes between the risk groups, adding a genomic dimension to the prognostic model.

Single-Cell Validation The ICDRS was also evaluated in single-cell transcriptomic datasets, confirming that the five-gene signature captures biologically meaningful variation at the level of individual cells within the tumor.

TL;DR: Genomic analyses show that ICDRS risk groups differ in mutation profiles and tumor mutational burden, with low-risk patients showing genomic features associated with better immunotherapy response.
Pages 10-11
From Computational Model to Tissue-Level Confirmation

Lab Validation Expression of the five ICDRS genes was measured in three NSCLC cell lines (H1975, A549, SK-LU-1) and one normal bronchial epithelial line using qRT-PCR, confirming that the predicted expression patterns held at the cellular level.

Immunohistochemistry IHC staining on formalin-fixed NSCLC tissue sections from 10 real patients confirmed protein-level expression of MMP14, one of the five key ICDRS genes, aligning computational predictions with histological evidence.

Nomogram Integration The ICDRS was integrated with clinical variables such as age, sex, and TNM stage into a nomogram for individualized survival prediction, with strong concordance index values and well-calibrated curves.

Drug Response Prediction The oncoPredict R package was used to estimate IC50 values for multiple drugs, confirming significant differences in predicted drug sensitivity between high-risk and low-risk patient groups.

TL;DR: Lab and tissue-level experiments confirmed the computational findings, and the ICDRS was combined with clinical factors into a practical nomogram for individual patient prognostication.
Pages 11-12
Limitations and the Path Toward Personalized NSCLC Treatment

Current Limitations The study is primarily retrospective and relies on publicly available datasets. Prospective clinical trials are needed to validate the ICDRS in real-world treatment decision-making contexts.

Small IHC Cohort Only 10 patients were included for immunohistochemical validation, and a larger cohort would strengthen the tissue-level confirmation of the five key genes.

Next Steps for ICD Modulation Beyond prediction, future work should investigate whether therapeutic modulation of ICD - through specific chemotherapy agents, radiation, or photodynamic therapy - can shift patients from the high-risk to the low-risk immune phenotype.

Broader Applications The multiomics and machine learning framework developed here could be extended to other cancer types, and future studies could explore how the ICDRS interacts with specific ICI agents to guide individual drug selection more precisely.

TL;DR: While promising, this ICDRS model requires prospective validation before clinical adoption, and future work should explore whether ICD-inducing therapies can actively convert immunologically cold NSCLC tumors into hot ones.
Citation: Open Access, 2025. Available at: PMC12432684.