Fig. 1: Training and validation of “HCC-detect” DNA methylation marker set.

A Heatmap and hierarchical clustering showing methylation levels of top 20 CpGs that are categorically different between noncancer liver samples (fibrosis) (n = 79) and HCC samples (n = 66) in the training cohort (GSE61258, GSE54503). B Scatterplot of “HCC-detect” methylation scores calculated for HCC samples (n = 66) and controls (n = 79) in the training cohort (p < 0.0001, Man-Whitney test, two tailed), The line at the median with 95% confidence interval is also shown in the plot. C ROC curve of “HCC-detect” methylation scores classifying blood and HCC samples from the training cohort. D Heatmap of methylation values for the four CpG sites included in “HCC-detect” in the validation cohort of blood (n = 968), heathy liver (n = 16), liver NAT (n = 116) and HCC samples (n = 739) from TCGA and GEO data repository (see Table 1 for details). E Scatterplot of “HCC-detect” Methylation score (each spot represents one sample) in healthy blood (n = 968), healthy liver tissue (n = 16), liver disease (n = 79), liver cirrhosis (n = 130), non-liver healthy tissues (n = 234), NAT to non-HCC tumors (n = 723), NAT to HCC (n = 116), 31 cancers in TCGA (n = 8754) and HCC (n = 739). The statistical analysis was performed using Kruskal-Wallis nonparametric ANOVA (two-sided) with Dunn’s multiple comparisons to compare each group to the HCC group. The statistical analysis revealed a significant difference between the HCC group and all other groups (F = 793, DF 11696; p < 0.0001 for all comparisons). The line at the median with 95% confidence interval is shown in the plot F ROC curve of “HCC-detect” methylation scores classifying healthy blood (968) and HCC samples (739) in the validation cohort. G. ROC curve of “HCC-detect” methylation scores classifying healthy tissues (2212) and HCC samples (739) in the validation cohort. H ROC curve of “HCC-detect” methylation scores classifying HCC samples (739) and 8754 samples from 31 different types of cancer (Supplementary Table 1). I The median +/− methylation scores for “HCC-detect” with 95% confidence intervals in HCC NATs (n = 50) and different stages of HCC in the validation cohort from TCGA (Stage A = 175, Stage B = 87, Stage C = 86, Stage D = 5). The statistical analysis was performed using Kruskal-Wallis nonparametric ANOVA (two-sided) with Dunn’s multiple comparisons test, with statistical significance indicated as **p < 0.001 and ****p < 0.0001. The line at the median with 95% confidence interval is shown in the plot. Source data are provided as a Source Data file.