Fig. 5

Validation using TCGA cohort. Note the insufficient data in upstream region beyond −1.5 kb of TSS due to lack of HM450K probes. a, b Scatter plots (left panels) separated by gene expression quartiles based on the expression levels in either tumor or normal samples (n = 3208, 3208, 3208, 3208 for Q1, Q2, Q3, Q4, respectively; Q4 is the highest expression) showing methylation ratio at HM450K probes located within (a) and outside (b) CpG islands. Right panels show the mean values. c Methylation-expression correlation density plot showing an increase of a density peak where DNA methylation and gene expression are significantly correlated (FDR q < 0.05). Plots of any correlation in all genes (n = 19004) in the TCGA validation data set, and negative correlation in CGI genes (n = 12832) and noCGI genes (n = 6172) are shown. d Significance of DNA methylation increase in tumor. The red line represents the observed AUC ratio (tumor/normal) at TSS ± 500 bp region. Gray bars represent the null distribution as calculated by multiple permutations (n = 1000). Results are also shown for CGI genes (n = 12873) among HPV-negative head and neck squamous cell carcinoma (e, HPV(-)HNSCC), colon adenocarcinoma (f, COAD), and breast invasive carcinoma (g, BRCA) available from TCGA