Fig. 4: Contributions of the tumor and cfDNA reference data.
From: Integrative modeling of tumor genomes and epigenomes for enhanced cancer diagnosis by cell-free DNA
A ROC-AUC and accuracy of the genome model depending on whether our reference cfDNA panel was used for cancer detection (left) and localization (right). B ROC-AUC and accuracy of the genome model according to which genomic regions were used for LVD feature acquisition in cancer detection (left) and localization (right). The total number of cancer type-specific high or low LMD regions are 264. The same number of random regions were selected for comparison with high or low LMD regions. C ROC-AUC and accuracy of the epigenome model depending on which NDRs were used for the construction of V-plot images for cancer detection (left) and localization (right). The same number of random regions were selected for comparison with tissue-specific NDRs. A–C Five-fold cross-validation was performed to examine the distribution of ROC-AUC and accuracy values. Each box indicates IQR and median, whiskers indicates 1.5× IQR, black dots indicates outlier. CD cancer detection, Localization cancer localization. Source data are provided as a Source Data file.
