Skip to main content
Fig. 5 | Genome Biology

Fig. 5

From: Systems-epigenomics inference of transcription factor activity implicates aryl-hydrocarbon-receptor inactivation as a key event in lung cancer development

Fig. 5

SEPIRA and LungNet predict preferential inactivation of lung-specific TFs during progression to LSCC, including LCIS. a–e RNA expression. a Heatmap of t-statistics of differential TF activity, as estimated using SEPIRA from a gene expression data matrix encompassing all major histological stages of lung carcinogenesis. N normal, H hyperplasia, M metaplasia, D dysplasia, LCIS lung carcinoma in situ, ILC invasive lung cancer (squamous). *TFs with significant changes in TF activity during disease progression. b Numbers of significantly deactivated (DN) and activated (UP) TFs in each disease stage relative to normal. c Boxplots of the t-statistics of differential activity between each disease stage and normal lung. P values are from a one-tailed Wilcoxon rank sum test, testing that the distribution of the differential activity values is < 0. c Scatterplot of t-statistics from a regression of TF-activity against disease stage (x-axis) against their significance level (-log10P, y-axis). d Boxplot of estimated TF-activity levels for TBX2 against disease stage. f DNAm. Left: Boxplots comparing the t-statistics of differential activity, estimating using SEPIRA on Illumina 450 k DNAm data, between 35 LCIS and 21 NADJ samples for the 38 LungNet TFs against a null model in which the targets of the 38 TFs were randomized among all possible targets (keeping the number of targets per TF fixed). P values above boxes represent the Wilcoxon rank sum test P values testing that the distribution of t-statistics is < 0. P value in-between boxes compares the distribution of t-statistics. Right: Density distribution of average t-statistics of differential activity obtained by performing 1000 randomizations of the targets (gray curve) against the observed average t-statistic of differential activity (red vertical line). None of the 1000 randomizations led to an average statistic lower than the observed (P < 0.001)

Back to article page