Due to insufficient early detection methods, ovarian cancer remains a leading cause of death among gynecological malignancies. Therefore, DNA methylation profiling in blood samples was investigated for its potential in detecting ovarian cancer in this study. Blood samples (n = 373) and tissue samples (n = 152) from healthy women, women with benign ovarian tumors, and women with malignant epithelial ovarian tumors were subjected to targeted bisulfite sequencing.
To forecast and categorize each blood sample as malignant or non-malignant, a supervised machine-learning system was constructed and cross-validated using the DNA methylation profiles of the training cohort (n = 178). A separate test cohort (n = 184) was used to assess the model. Further, there were 1,272 differentially methylated sites were found after comparing DNA methylation profiles of benign and malignant tumor samples; 49.4% of these regions were hypermethylated, and 50.6% were hypomethylated. The area under the curve for the model in the training dataset was 0.94 when researchers performed 5-fold cross-validation.
The algorithm correctly identified benign tumors in the test dataset in 93.5% of women with benign tumors (n = 46) and 96.2% of healthy women (n = 53). The model correctly identified malignancy in 44.4% of stage I-II (n = 9), 86.4% of stage III (n = 59), 100% of stage IV (n = 6), and 81.8% of stage unknown (n = 11) tumors in patients with malignant tumors. The overall prediction accuracy of the model was 89.5%. These results show that DNA methylation profiling in blood could be used clinically to screen for ovarian cancer.