Fig. 6: Interpretation of the epigenome model.
From: Integrative modeling of tumor genomes and epigenomes for enhanced cancer diagnosis by cell-free DNA
A Attribution values mapped to the tissue-specific V-plots of the epigenome model for cancer detection on the MGI training cohort. The average attribution values across the tissue-specific V-plots are compared for cancer samples (left) and normal samples (right). B Distribution of the normalized attribution values according to fragment size. C Attribution values mapped to the tissue-specific V-plots of the epigenome model for tissue-of-origin localization on the MGI training cohort. The average attribution values across the tissue-specific V-plots are shown for each cancer type. D Attribution values of the epigenome model for cancer localization on the MGI training cohort, as averaged across the NDRs of the matching (red) versus unmatching (blue) cancer types, and plotted according to the distance from the NDR midpoints. E The same plot as D broken down by cancer types. Attribution values were smoothed using lowess regression. F Clustered heatmaps of normalized attribution values in the epigenome model for (left) cancer detection and (right) cancer localization across normal samples and patient samples. The cancer types commonly present in the MGI training cohort and Illumina training cohort were included. Only the samples with a test prediction score above the threshold were included. The attribution values of all epigenome model features were normalized as the z score. Source data are provided as a Source Data file.
