Harnessing Machine Learning and Multiomics to Construct a Tumor-Specific T Cell Signature for Prognostic Assessment and Precision Medicine in Lung Adenocarcinoma

Abstract

Background: T cells are pivotal in mediating antitumor immunity in lung adenocarcinoma (LUAD). In this study, we aimed to profile T cell-related gene (TRG) expression and develop a prognostic indicator to identify patients with LUAD who may derive greater benefit from immunotherapy.

Patients and methods: Transcriptomic and clinical data of patients with LUAD were sourced from The Cancer Genome Atlas and Gene Expression Omnibus databases. The prognostic relevance of tumor-infiltrating T cells was assessed, and TRGs were further pinpointed through single-cell RNA-seq (scRNA-seq) analysis. Weighted gene coexpression network analysis identified LUAD-specific modules. A T cell-related gene prognostic indicator (TRGPI) was subsequently developed using a machine learning framework, with the RSF + Ridge model chosen on the basis of cross-cohort performance. We further employed spatial transcriptomics to evaluate the most impactful prognostic TRG, providing spatial context to its expression patterns.

Results: Increased T cell infiltration correlated with improved survival outcomes in LUAD. The TRGPI, derived from both scRNA-seq and bulk transcriptomic data, demonstrated robust prognostic and predictive capabilities across multiple cohorts. Patients with a low TRGPI exhibited enhanced overall survival, more active immune and antibacterial pathways, a higher tumor mutation burden, and more favorable predicted responses to immunotherapy. TPI1 was identified as the most impactful prognostic TRG, and spatial transcriptomics analysis and functional assays further established the oncogenic role of TPI1 in LUAD.

Conclusions: This study developed a novel, robust TRGPI that accurately predicts patient prognosis and immunotherapy responses in LUAD, providing a valuable tool for precision medicine and personalized treatment strategies.

Keywords: Immunotherapy; Lung adenocarcinoma; Machine learning; Multiomics; Prognostic indictor.