Fig. 3

Fragmentation profiles from cfDNA 5hmC data for detection of lung cancer. A cfDNA fragment lengths distribution of healthy individuals (n = 189, grey) and patients with lung cancer (n = 157, blue) based on 4 bp windows. Patterns of 110 bp–140 bp based on 1 bp windows were shown in gray box. B Pearson correlations are depicted with box plots to show minimum, 25th percentile, median, 75th percentile, and maximum values. P value = 6.34E−19. C Genome-wide cfDNA fragmentation profiles (defined as the ratio of short to long fragments) from 346 cfDNA 5hmC data are shown in 5-Mb bins for 189 healthy individuals (top) and 157 patients with lung cancer (bottom). D ROC curves of fragmentation model for lung cancer patients and healthy controls in training set (black), validation set (red) and external validation set (green). E Boxplot of the fragment wd-score calculated with fragmentation model for lung cancer samples and healthy controls (training set P value = 1.52E−37, validation set P value = 2.29E−14). F Boxplot of the fragment wd-score calculated with fragmentation model for healthy controls, stage I lung cancer samples and stage II–IV lung cancer samples. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 1e − 5, Wilcoxon test