Fig. 6

Multimodal epigenetic analysis of cfDNA TAPS improves the performance of cancer detection model. A Genome browser tracks showing sequencing signals of cfDNA TAPS of controls (cfTAPS of healthy controls) and nucleosome calls generated by ENCODE project with accession number ENCSR000CXP (Lymphoblastoid cell MNase-seq). Sequencing signals from cfDNA TAPS are calculated by DANPOS2. B Genomic regions showing nucleosome occupancy changes between HCC (left panel) or PDAC (right panel) and control samples. Nucleosome occupancy is calculated by DANPOS2. The top panel shows tracks of regions surrounding TSSs, and bottom panel shows regions surrounding polyadenylation sites. For each panel, two example cancer and control samples are displayed. The blue boxes show the gene bodies with white arrows indicating the transcription directions. C, D ROC curves showing the performances of two-class models which distinguish HCC (C) or PDAC (D) from control samples. The results from 51 iterations for (C) and 53 iterations for (D) of leave-one-out cross-validation analysis were shown. E Bar plot showing the overall accuracy of three-class models which distinguish HCC, PDAC, and control samples. The results from 74 iterations of leave-one-out cross-validation analysis are shown. Methylation, methylation ratio of promoter and enhancer regions; Occupancy, nucleosome occupancy of 1 kb regions surrounding TSSs and polyadenylation sites; WPS, WPS of 1 kb regions surrounding TSSs and polyadenylation sites; Multimodal, the combination of all three types of features