Integrated single-cell and spatial mapping coupled with machine learning unveils core stemness landscapes and regulatory drivers in triple-negative breast cancer.

Triple-negative breast cancer (TNBC) exhibits pronounced intratumoral heterogeneity, and cancer stem cells (CSCs) are thought to play a pivotal role in this process. However, the molecular regulatory mechanisms linking CSC-associated stemness features to tumor progression remain insufficiently elucidated. We integrated single-cell RNA sequencing (scRNA-seq), spatial transcriptomics, and bulk transcriptomic data to identify high-stemness cell populations using the inferCNV and CytoTRACE algorithms. Stemness-related genes were evaluated for feature importance through hdWGCNA combined with machine learning approaches, and an XGBoost-based risk prediction model was constructed. Cellular differentiation trajectories were inferred using Monocle3 and scTour, while the effects of core genes on stemness pathways and malignant biological behaviors were assessed via CellChat analysis, SHAP attribution, and scTenifoldKnk-based virtual knockdown experiments. We successfully established a predictive model comprising five core stemness-related genes (CALD1, ANP32B, FIS1, CD82, and APLP2), with the high-stemness score group exhibiting poorer prognosis and enhanced immune evasion. Trajectory analysis confirmed that the high-stemness subpopulation resided at the initiation stage of differentiation. Enrichment analyses revealed highly active Notch signaling communication, and virtual knockdown of hub genes effectively suppressed the expression of stemness markers such as NOTCH1. In addition, drug sensitivity analysis identified BI.2536 and related compounds as exhibiting higher therapeutic sensitivity in the high-risk group. Our predictive model offers a novel perspective on the stemness landscape of TNBC. These core genes play key roles in maintaining stemness and also serve as potential molecular targets for personalized therapies aimed at TNBC stem-like cells.