Abstract
BACKGROUND: Head and neck squamous cell carcinoma (HNSCC) demonstrates significant heterogeneity, necessitating improved molecular classification for precision treatment. METHODS: We integrated single-cell and bulk RNA sequencing data from 59,376 cells across ten datasets using Scissor and scSTAR packages. Molecular subtyping was performed through ssGSEA and WGCNA analysis, with immune infiltration evaluated using CIBERSORT. We developed a machine learning-based risk prediction model using 54 algorithms. RESULTS: We identified three molecular subtypes with distinct prognostic implications, showing significant survival differences across independent datasets (TCGA-HNSCC, P < 0.0001; GSE65858, P = 0.018). The C3 subtype showed enhanced immunotherapy response potential, while C2 exhibited the highest genomic alteration rate (97.06%) and TP53 mutations (80%). Macrophages emerged as key players in intercellular communication networks. Our risk prediction model demonstrated robust performance across four validation cohorts. CONCLUSION: This molecular subtyping framework provides valuable insights for patient stratification and personalized therapeutic strategies in HNSCC, potentially improving clinical outcomes through precise treatment selection.