A Machine Learning-Derived Taurine Metabolism Signature Predicts Prognosis and Immune Landscape in Lung Adenocarcinoma via Integrative Single-Cell Analysis.

Background: Lung adenocarcinoma (LUAD) represents a biologically diverse tumor type, often associated with unfavorable prognosis and unsatisfactory therapeutic outcomes. Over the past few years, increasing attention has been given to metabolic alterations as key contributors to cancer development. Nevertheless, the specific contribution of taurine-related metabolic pathways in LUAD remains unclear. By constructing a model from taurine metabolism-associated genes, we aimed to elucidate its mechanistic basis and evaluate its relevance to clinical outcomes. Methods: Transcriptomic profiles and clinical annotations from TCGA along with five LUAD datasets from the GEO repository were comprehensively integrated. A risk score indicative of taurine metabolism-associated signature (taurine-related signature [TRS]) was constructed by integrating LASSO regression, stepwise Cox modeling, and SuperPC algorithm. Its predictive capability was systematically evaluated using Kaplan-Meier survival analysis, ROC curves, and DCA. To further investigate the relationship between TRS and both cellular heterogeneity and tumor microenvironmental context, single-cell RNA-seq data were integrated into the analysis. Moreover, the tumorigenic role of the hub gene KIF2C was experimentally validated via in vitro functional assays. Results: The TRS signature was validated for its predictive relevance using six LUAD datasets from independent sources, showing its ability to categorize patients based on survival variations. Elevated TRS levels were strongly linked to increased tumor cell proliferation, immune evasion characteristics, and impaired response to immunotherapy. Findings from single-cell RNA sequencing indicated that epithelial subpopulations with higher TRS expression displayed intensified metabolic activity and reduced antigen presentation efficiency. In particular, KIF2C-a key component of the TRS gene set-was highly expressed in LUAD tissues and associated with less favorable prognostic profiles. Functional silencing of KIF2C led to decreased proliferation and invasive capacity of LUAD cells, supporting its potential role as a tumor-promoting factor. Conclusion: We established a taurine metabolism-related prognostic model (TRS) and investigated its function in LUAD by combining transcriptomic information from both bulk tissues and single-cell datasets. The TRS effectively categorizes patients according to their prognosis and reflects immune-related features. KIF2C, identified as a crucial gene within the TRS, could be explored as a potential therapeutic candidate, offering insights into LUAD metabolism and informing the development of metabolism-based therapeutic strategies.