Fig. 2
From: Modelling the conditional regulatory activity of methylated and bivalent promoters
Analysis of predictive models of genome-wide transcript abundance for a H1-hESC, b GM12878 and c K562 cell lines, constructed from H2A.Z, H3K4me3, H3K27me3 and H3K9me3 histone scores. Each cell line demonstrates the following: (top) the distribution of arsinh-transformed RPKM-normalised transcript abundance derived from RNA-seq data; (middle) predicted-versus-measured transcript abundance for the linear regression model, with performance quantified as an adjusted R 2 score; and (bottom) the data-derived putative regulatory roles of each histone modification, with positive/negative loadings suggesting activator/repressor roles, respectively. Of particular interest is the latent signature of DNA methylation-associated gene silencing, with GM12878 and K562 exhibiting a higher proportion of near-zero expression genes and strikingly stronger regulatory signal for H3K9me3 (implicated in DNA de novo methyltransferase activity)