Abstract
The Head and Neck Squamous Cell Carcinoma (HNSCC), arising from the mucosal epithelium of the oral cavity, pharynx, and larynx, continues to represent a major worldwide health burden due to its high mortality rates and late-stage diagnosis. A contribution of this study is the focus on the heterogeneity of CAFs, which directly impacts therapeutic response and resistance. To address this, we applied an AI-driven, multi-omics integration strategy to elucidate CAF-mediated mechanisms in HNSCC progression and therapy. Bulk transcriptomic data from Gene Expression Omnibus (GEO) were intersected with curated CAF gene sets to identify CAF-related differentially expressed genes (CAFs-DEGs). To create a fibroblast-associated prognosis signature, a machine learning-based LASSO-Cox regression model has been developed using the TCGA-HNSCC cohort. Prognostic performance was validated through Kaplan-Meier survival analysis, time-dependent ROC, nomogram, Decision Curve Analysis (DCA), and calibration curves. To provide mechanistic insights, immune infiltration profiling, checkpoint correlations, single-cell expression mapping, tumor mutational burden (TMB), microsatellite instability (MSI), and DNA methylation analyses were performed. Furthermore, therapeutic vulnerabilities were explored by integrating drug sensitivity prediction, AI-assisted cMAP screening, and molecular docking validation, which identified Epothilone B as a promising agent targeting HBEGF. Overall, this research shows that understanding the heterogeneity of CAFs with AI-enabled multi-omics modeling can reveal prognostic biomarkers and therapeutic targets for overcoming resistance, with the ultimate goal of improving precision oncology for HNSCC.