The high lethal rate of pancreatic cancer is partly due to a lack of efficient biomarkers for screening and early diagnosis. We attempted to develop effective and noninvasive methods using 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC) markers from circulating cell-free DNA (cfDNA) for the detection of pancreatic ductal adenocarcinoma (PDAC). A 24-feature 5mC model that can accurately discriminate PDAC from healthy controls (area under the curve (AUC) = 0.977, sensitivity = 0.824, specificity = 1) and a 5hmC prediction model with 27 features demonstrated excellent detection power in two distinct validation sets (AUC = 0.992 and 0.960, sensitivity = 0.786 and 0.857, specificity = 1 and 0.993). The 51-feature model combining 5mC and 5hmC markers outperformed both of the individual models, with an AUC of 0.997 (sensitivity = 0.938, specificity = 0.955) and particularly an improvement in the prediction sensitivity of PDAC. In addition, the weighted diagnosis score (wd-score) calculated with the 5hmC model can distinguish stage I patients from stage II–IV patients. Both 5mC and 5hmC biomarkers in cfDNA are effective in PDAC detection, and the 5mC-5hmC integrated model significantly improve the detection sensitivity.

中文翻译：

---

循环游离 DNA 中的整合表观遗传生物标志物作为胰腺癌的强大分类器。


胰腺癌的高致死率部分是由于缺乏有效的生物标志物进行筛查和早期诊断。我们尝试使用循环游离 DNA (cfDNA) 中的 5-甲基胞嘧啶 (5mC) 和 5-羟甲基胞嘧啶 (5hmC) 标记物开发有效且非侵入性的方法来检测胰腺导管腺癌 (PDAC)。具有 24 个特征的 5mC 模型可以准确地区分 PDAC 与健康对照（曲线下面积 (AUC) = 0.977，灵敏度 = 0.824，特异性 = 1），而具有 27 个特征的 5hmC 预测模型在两个不同的验证集中表现出出色的检测能力（AUC = 0.992 和 0.960，敏感性 = 0.786 和 0.857，特异性 = 1 和 0.993）。结合 5mC 和 5hmC 标记的 51 个特征模型的性能优于两个单独的模型，AUC 为 0.997（灵敏度 = 0.938，特异性 = 0.955），尤其是 PDAC 的预测灵敏度有所提高。此外，用5hmC模型计算的加权诊断评分（wd-score）可以区分I期患者和II-IV期患者。 cfDNA中的5mC和5hmC生物标志物在PDAC检测中均有效，并且5mC-5hmC集成模型显着提高了检测灵敏度。