EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Conserved Fibrinogen and Immune Evasion Gene Signature Predicts Mortality Across Lung Cancer Histological Subtypes: An Interpretable Machine Learning Discovery Study

Simbarashe Magwenzi

Journal of Precision Medicine and Artificial Intelligence · 09.06.2026

DOI: 10.64949/sfa07913 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Precision Medicine and Artificial Intelligence, online erschienen am 09.06.2026. Verfasst von Simbarashe Magwenzi. Die Arbeit nennt 24 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This article is a preprint and has not yet been peer-reviewed. Not for clinical use. BackgroundTNM staging systematically misclassifies patients with favourable outcomes as high-risk, with direct consequences for treatment decisions. Whether the molecular drivers of lung cancer mortality are conserved across histological subtypes remains largely uncharacterised.MethodsAn XGBoost classifier was trained on 104 features (100 gene expression probes and four clinical variables) in 440 lung adenocarcinoma (LUAD) patients (GSE68465). A three-model ablation study quantified the independent contribution of gene expression over clinical staging alone. The trained model was applied without modification to an independent cohort (GSE30219, n=287), stratified into adenocarcinoma (n=85) and mixed-histology (n=202) subsets. SHapley Additive exPlanations (SHAP), Kaplan-Meier survival analysis, and decision curve analysis assessed biological and clinical utility.ResultsClinical staging alone misclassified 49% of surviving patients as high-risk (specificity=0.51). Adding gene expression reduced this to 29% (specificity=0.71), corresponding to approximately 93 fewer incorrect high-risk designations per 1,000 patients. The full model achieved AUC=0.73 in the training test set, with stable external validation in the LUAD (AUC=0.71) and mixed-histology (AUC=0.69) subsets. When applied to the mixed-histology subset, the LUAD-trained model maintained near-identical biological feature hierarchies, with 12 of 14 top SHAP features shared across all three evaluation cohorts. The dominant signal in both external validation cohorts was the fibrinogen chain gene pair FGG and FGA, outranking pathological nodal stage and implicating coagulation-mediated tumour immune exclusion as a conserved mortality mechanism. Kynureninase (KYNU), a key mediator of IDO/TDO-mediated immune evasion, was independently recovered as a top predictor. Risk stratification significantly separated disease-free survival in both validation subsets (LUAD: log-rank p=0.044; mixed-histology: log-rank p=0.011).ConclusionGene expression substantially improves survivor identification over clinical staging alone and reveals a conserved molecular signature of lung cancer mortality transcending histological boundaries. The dominance of fibrinogen and kynurenine pathway components across multiple histotypes suggests histotype-independent mechanisms of aggressiveness with direct implications for prognostic stratification and therapeutic targeting.

Zitieren

APA

Magwenzi, S. (2026). A Conserved Fibrinogen and Immune Evasion Gene Signature Predicts Mortality Across Lung Cancer Histological Subtypes: An Interpretable Machine Learning Discovery Study. Journal of Precision Medicine and Artificial Intelligence. https://doi.org/10.64949/sfa07913

Vancouver

Magwenzi S. A Conserved Fibrinogen and Immune Evasion Gene Signature Predicts Mortality Across Lung Cancer Histological Subtypes: An Interpretable Machine Learning Discovery Study. Journal of Precision Medicine and Artificial Intelligence. 2026. doi:10.64949/sfa07913

BibTeX
@article{Magwenzi2026,
  author = {Magwenzi, Simbarashe},
  title = {A Conserved Fibrinogen and Immune Evasion Gene Signature Predicts Mortality Across Lung Cancer Histological Subtypes: An Interpretable Machine Learning Discovery Study},
  journal = {Journal of Precision Medicine and Artificial Intelligence},
  year = {2026},
  doi = {10.64949/sfa07913}
}
Kostenlos starten →