Fig. 4

A linear regression model accurately predicts cell-type-specific promoter activity. a Density scatterplot comparing RNA-seq values for single-promoter genes to K562 promoter scores predicted by the model trained on K562 data. b Scatterplot of promoter chromatin scores for single-promoter genes in H1 and K562 cells. Colored dots indicate that the promoter scores are ≥ twofold enriched in either H1 cells (cyan) or K562 cells (magenta). c Scatterplot of promoter scores that are ≥ twofold enriched in either H1 cells (cyan) or K562 cells (magenta) mapped onto their corresponding RNA-seq values. Blue dotted lines indicate the twofold difference cutoff. d Scaled Venn diagram showing the overlap between genes called as cell type specific according to their promoter scores, or according to their RNA expression values. Genes predicted to have contradictory cell type specificities according to promoter activity modeling versus RNA-seq are indicated (scaled black circle). e Scatterplot comparing the H1 and K562 scores of all promoters separated by ≥ 2 kb. Master regulators of H1 and K562 cell identities are indicated as colored circles. Both OTX2 and TAL1 have two promoters that can be distinguished