Ferroptosis-Linked Six-Gene Panel Enables Machine Learning-Assisted Diagnosis and Therapeutic Guidance in Lung Adenocarcinoma.

Abstract

Lung adenocarcinoma (LUAD) remains the most common subtype of non-small-cell lung cancer and a major cause of cancer mortality, with many patients lacking actionable mutations or durable responses to targeted or immune therapies. Here, we report an integrative analysis of TCGA LUAD transcriptomes (n = 598) seeded from a curated ferroptosis gene catalogue, yielding a compact six-gene signature (AQP4, CDCA3, HJURP, KIF20A, PLK1, UHRF1) with diagnostic, prognostic, and therapeutic relevance. The signature was consistently dysregulated in tumours versus normal lung and stratified patients into high- and low-risk groups with distinct survival outcomes (log-rank p < 0.0001), outperforming conventional staging when incorporated into multivariable models. Across ten machine learning algorithms, the panel achieved near-perfect tumour-normal classification (AUC 0.99-1.00), highlighting its translational potential for early detection. Functional analyses linked the signature to cell-cycle, angiogenic, and immune modulation, while exploratory drug-gene correlations identified PLK1 and other candidates as potential therapeutic targets. Together, these findings establish a biologically anchored six-gene panel that complements existing mutation-based classifiers and provides a framework for advancing diagnostic precision, prognostic refinement, and biomarker-guided therapeutic strategies in LUAD.