The construction and preliminary validation of methylation haplotype load markers of multiple genes for cervical cancer screenings

Abstract Background At present, there are certain deficiencies in cervical cancer screening methods. Consequently, we aimed to establish a prediction model for cervical precancerous lesions utilizing DNA methylation haplotype load (MHL) markers. Methods Three machine learning models, namely, a random forest model (RF), a support vector machine model (SVM), and a naive Bayes model (NB), were developed based on the performance of 47 regions in 52 HR-HPV + cervical cytology specimens. Additionally, external validation of the three models was conducted using 101 HR-HPV + cervical cytological samples. Results From the initial 52 samples, we derived three models with respective area under the curve (AUC) values of 0.864, 0.867 and 0.847. Subsequently, in the validation phase involving 101 samples, the RF model demonstrated superior performance compared to the E6 protein detection model and p16Ki67 model in predicting cervical intraepithelial neoplasia grade 2 (CIN2) lesions and above. Compared with the human papillomavirus (HPV) combined with cytology triage model, the RF prediction model had a PPV of 100% and an NPV of 95.15% in the HPV16/18-positive subgroup. Within the other high-risk-HPV-positive subgroups, the PPV and NPV were 41.67% and 92.36%, respectively. Conclusions Our findings suggest that the methylation haplotype load markers of multiple genes offer clear advantages in screening CIN2 lesions and above in cervical cancer. Furthermore, enhancing the methylation detection method has the potential to further improve the NPV of the prediction model.