Development of a machine learning-derived dendritic cell signature for prognostic stratification in lung adenocarcinoma.

Background: Lung adenocarcinoma (LUAD), the most common histological subtype of lung cancer, demonstrates significant intertumoral heterogeneity. While dendritic cells (DCs) are essential mediators of antitumor immunity, their transcriptional diversity and prognostic value in LUAD remain underexplored. Methods: We constructed a cellular atlas by integrating single-cell RNA sequencing (scRNA-seq) data from LUAD and normal tissues, emphasizing dendritic cells. High-dimensional weighted gene co-expression network analysis (hdWGCNA) and pseudotime analysis were utilized to identify functional modules and lineage trajectories. A dendritic cell-related signature (DCRS) was constructed using multiple machine learning algorithms (Lasso-Cox, RSF, CoxBoost, Stepwise-Cox), and its prognostic performance was validated in seven external cohorts. Immune landscape, genomic instability, drug sensitivity, and immunotherapy response were further analyzed. The functional role of PLEK2, a DCRS hub gene, was validated in clinical samples and LUAD cell lines. Results: We identified six DC clusters with distinct developmental states and transcriptional programs. The M2 module was enriched in prognostically relevant clusters and used to derive the DCRS. Patients in the high-DCRS group exhibited worse prognosis, lower immune infiltration, higher chromosomal instability and tumor mutation burden, and reduced responsiveness to immunotherapy. Drug sensitivity analysis revealed that the low-DCRS group was more responsive to multiple chemotherapeutic agents. Functional validation confirmed that Conclusion: We established a novel DCRS with robust prognostic and predictive value in LUAD. This work highlights the pivotal role of dendritic cell programs in shaping the tumor microenvironment and provides potential targets for improving precision immunotherapy.