Abstract：
In the Circulating Cell-free Genome Atlas (NCT02889978) substudy 1, we evaluate several approaches for a circulating cell-free DNA (cfDNA)-based multi-cancer early detection (MCED) test by defining clinical limit of detection (LOD) based on circulating tumor allele fraction (cTAF), enabling performance comparisons. Among 10 machine-learning classifiers trained on the same samples and independently validated, when evaluated at 98% specificity, those using whole-genome (WG) methylation, single nucleotide variants with paired white blood cell background removal, and combined scores from classifiers evaluated in this study show the highest cancer signal detection sensitivities. Compared with clinical stage and tumor type, cTAF is a more significant predictor of classifier performance and may more closely reflect tumor biology. Clinical LODs mirror relative sensitivities for all approaches. The WG methylation feature best predicts cancer signal origin. WG methylation is the most promising technology for MCED and informs development of a targeted methylation MCED test.
摘要：
在循环无细胞基因组图谱（NCT02889978）子研究1中，我们通过定义基于循环肿瘤等位基因分数（cTAF）的临床检测限（LOD）来评估基于循环无细胞DNA（cfDNA）的多癌早期检测（MCED）测试的几种方法。启用性能比较。在对相同样本进行训练并独立验证的10个机器学习分类器中，当在98%的特异性进行评估时，那些使用全基因组(WG)甲基化的人，具有配对白细胞背景去除的单核苷酸变体，在这项研究中评估的分类器的组合得分显示出最高的癌症信号检测灵敏度。与临床分期和肿瘤类型相比，cTAF是分类器性能的更显著的预测因子,并且可以更密切地反映肿瘤生物学。临床LOD反映了所有方法的相对敏感性。WG甲基化特征最好地预测癌症信号起源。WG甲基化是MCED最有前途的技术，并告知靶向甲基化MCED测试的发展。
