A DNA methylation-based machine learning model for early and accurate diagnosis of cervical HSIL+ lesions

Abstract Current cervical cancer screening methods lack accuracy in early diagnosis and risk prediction. We developed a DNA methylation-based diagnostic model for cervical high-grade squamous intraepithelial lesions or more severe lesions (HSIL+). This study systematically collected 172 liquid-based cytology samples from patients with positive human papillomavirus (HPV) test results. Bisulfite conversion-based next-generation sequencing (NGS) methylation sequencing technology was employed to quantitatively assess the methylation levels of consecutive CpG sites within specific segments in MIR9-3HG, TERT, GATA3, and CDKN2A genes across various grades of cervical lesions. Machine learning algorithms (LASSO regression, random forest, and support vector machine [SVM]) identified methylated characteristic CpG sites.The methylation levels of the four genes detected in the HSIL/ cervical squamous cell carcinoma(SCC) group were significantly higher than those in the Negative for Intraepithelial Lesion or Malignancy(NILM)/ low - grade squamous intraepithelial lesions (LSIL) group. Receiver Operating Characteristic (ROC) curve analysis showed that the areas under the curve (AUCs) for CDKN2A, MIR9-3HG, GATA3 and TERT in diagnosing HSIL+ were 0.880 (95% CI: 0.824–0.937), 0.779 (95% CI: 0.704–0.854), 0.769 (95% CI: 0.684–0.855) and 0.713 (95% CI: 0.627–0.800), respectively. The CpG methylation sites selected by the support vector machine (SVM), random forest algorithm and LASSO regression analysis were further cross-validated by Venn diagram. Finally, four CpG sites (all located in CDKN2A) were selected to successfully construct an efficient diagnostic model for HSIL+, with a sensitivity of 0.704 and a specificity as high as 0.929. The diagnostic model constructed in this study can accurately diagnose HSIL + of the cervix at an early stage, and it has significant clinical application value.