Abstract
Radiotherapy is the cornerstone of treatment for cervical cancer, yet the variability of patient response demands a deeper understanding of the molecular determinants of radioresistance. In this study, we investigated the molecular and cellular mechanisms of radioresistance in cervical cancer through a comprehensive multi-omics and machine learning approach. We downloaded and processed transcriptome sequencing, methylation and single-cell sequencing data from the TCGA and GEO databases. Differential gene and methylation analyses were performed to identify radioresistance-related markers. Single-cell data were processed using Seurat and annotated using CellTypist. Prognostic models were constructed and validated through downscaling, cell scoring, trajectory analysis and machine learning. Additionally, immune infiltration and drug sensitivity analyses were conducted. The differential analysis identified 845 up-regulated and 460 down-regulated genes associated with radioresistance. The methylation analysis identified 3042 down-regulated and 158 up-regulated gene loci. Single-cell sequencing revealed 43,475 cells and 13 cell types, with aneuploid cells predominantly present in epithelial cells. Cell scoring highlighted dispersed immune cells, with monocytes, ILCs, and T cells being the most relevant to radiotherapy resistance. The machine learning approach constructed a robust prognostic model using Cox regression and validated it on multiple datasets. The prognostic model demonstrated good predictive ability in assessing radiotherapy efficacy and immune infiltration. Drug screening identified several potential therapeutic candidates with high sensitivity for high-risk patients. This study provides a comprehensive multi-omics analysis and machine learning framework for identifying and validating molecular markers and prognostic models associated with radioresistance in cervical cancer, providing insights for personalized treatment strategies.