Abstract
Tumor stemness represents a key biological process that drives tumor progression and therapeutic resistance across various cancer types. To systematically elucidate the regulatory roles of long non-coding RNAs (lncRNAs) in this process, we integrated bulk transcriptomic data from The Cancer Genome Atlas (TCGA) with publicly available pan-cancer single-cell transcriptomic atlases. Using machine-learning-based stemness metrics, we successfully quantified stemness features and identified unique lncRNA gene sets for each cancer type at the bulk data level. The high-stemness subtype exhibited enhanced proliferation, an immunosuppressive microenvironment, and profound metabolic reprogramming. Based on these findings, we constructed a robust prognostic model with remarkable predictive performance across multiple cancer types. At the single-cell resolution, we reconstructed the dynamic trajectory of stemness evolution, uncovering distinctive metabolic and cell-communication patterns within cancer stem cells (CSCs). This multi-scale analysis consistently nominated a core set of regulatory lncRNAs, including NEAT1 and MALAT1. Our work not only nominates potential targets for stemness-directed therapy but also provides a comprehensive framework for understanding lncRNA-driven mechanisms of cancer aggressiveness and resistance.