Supplementary Table 3: Investigating and modeling the differential DNA methylation for early lung adenocarcinoma diagnosis

Background: Aberrant DNA methylations serve as rich sources of diagnostic biomarkers, but a further improvement in their accuracy and clinical utility is warranted. Methods: Large panel bisulfite sequencing were performed on paired normal and stage I/IV tumors from 226 lung adenocarcinoma cancer patients to characterize the differentially methylated regions (DMRs). Results: Random forest model achieved high prediction accuracy (sensitivity 96% and specificity 97.56%) to separate normal controls from both early and advanced cancer samples, which is superior to most previous prediction models tested in lung adenocarcinoma. Conclusion: Our results suggest that combining the random forest model with targeted bisulfite sequencing have great clinical potentials to accurately predict and early diagnose lung adenocarcinoma during cancer screening.

研究背景：异常DNA甲基化是诊断生物标志物（biomarker）的优质来源，但仍需进一步提升其诊断准确性与临床应用价值。研究方法：本研究对226例肺腺癌患者的配对正常组织与I/IV期肿瘤组织开展大Panel亚硫酸氢盐测序，以表征差异甲基化区域（differentially methylated regions, DMRs）。研究结果：随机森林（Random Forest）模型实现了较高的预测准确率（灵敏度96%、特异度97.56%），可有效区分正常对照与早期及晚期癌症样本，其性能优于多数已在肺腺癌中验证的既往预测模型。研究结论：本研究结果显示，将随机森林模型与靶向亚硫酸氢盐测序相结合，在癌症筛查场景中具备精准预测并早期诊断肺腺癌的巨大临床应用潜力。