Abstract
BACKGROUND: Breast cancer is the most common malignancy in women globally, with significant heterogeneity affecting prognosis and treatment. RNA-binding proteins play vital roles in tumor progression, yet their prognostic potential remains unclear. This study introduces an Artificial Intelligence-Assisted RBP Signature (AIRS) model to improve prognostic accuracy and guide personalized treatment. METHODS: Data from 14 BC cohorts (9,000+ patients) were analyzed using 108 machine learning model combinations. The AIRS model, built on three key RBP genes (PGK1, MPHOSPH10, MAP2K6), stratified patients into high- and low-risk groups. Genomic alterations, single-cell transcriptomics, tumor microenvironment characteristics, and drug sensitivity were assessed to uncover AIRS-associated mechanisms. RESULTS: The AIRS model demonstrated superior prognostic performance, surpassing 106 established signatures. High AIRS scores correlated with elevated tumor mutational burden, specific copy number alterations, and an immune-suppressive TME. Single-cell analysis revealed functional heterogeneity in epithelial cells, linking high AIRS scores to pathways like transcription factor binding. Regulatory network analysis identified key transcription factors such as MYC. Low AIRS scores predicted better responses to immune checkpoint inhibitors, while drug sensitivity analysis highlighted panobinostat and paclitaxel as potential therapies for high-risk patients. CONCLUSIONS: The AIRS model offers a robust tool for BC prognosis and treatment stratification, integrating genomic, transcriptomic, and single-cell data. It provides actionable insights for personalized therapy, paving the way for improved clinical outcomes. Future studies should validate findings across diverse populations and expand functional analyses.