Fig. 4

Fragment length profiles of LB-unique mutations could distinguish liquid biopsy samples of patients with early-stage HCC from healthy individuals. A Comparison of ratios of short (< 150 bp) to long (> 150 bp) reads bearing LB-unique mutations (feature 1) between PwHCC and healthy individuals in the discovery cohort. Box plots include the median line, p-value estimated by the one-tailed Mann–Whitney U test. B Comparison of PHRED-scaled p-values obtained from enrichment analysis of fragment length distribution in specific ranges (feature 2, Materials and methods sections) of all fragments bearing LB-unique mutations from PwHCC with that of healthy individuals. Box plots include the median line, p-value was using the one-tailed Mann–Whitney U test. C Principal component analysis of \(\lambda \left(s\right)\) over a sliding and non-overlapping window of 10 consecutive values of \(s\), \(s\in \{100, 101,..., 250\}\), which yields a 15-dimensional feature vector (feature 3) in PwHCC and healthy individuals. D ROC curves showing the classification power of individual features and their combinations. E Box plot comparing the cancer prediction scores of patients with different tumor stages including stage I (n = 9), stage II (n-29) and stage III HCC (n = 3). F ROC curve showing the performance of models built by combining the three signatures of fragment length of LB-unique mutations in an independent validation cohort. bp: base pair