Fig. 2

Distribution of epigenetic marks at p65 summits and p65 recruitment by combinations of frequently co-occurring epigenetic marks. a Enrichment profiles of all 8 epigenetic marks around p65 peak summits at accessible and inaccessible sites. Enrichment levels should be compared for a single mark between the two classes of sites, but not between different marks for a single class, as magnitudes of ChIP-seq peaks are, in general, not comparable. b Densities of epigenetic marks at p65 peaks within inaccessible regions compared to accessible stage 1 (S2) TSS-distal regulatory sites, bound by p65 (+) or not (−) (see Methods). We applied non-negative matrix factorization (NMF) to matrices of scaled levels of epigenetic marks separately for the three classes of regulatory sites (TSS-proximal, TSS-distal stage 1, TSS-distal stage 2). For each class the method returned four combinations (codes) of strongly associated epigenetic marks. Next, we trained logistic regression models to predict p65 binding from locus-specific weights for each code and NF-κB motif occurrence (mot). c Within each code, the loadings of marks are tied, which reflects their most frequent relative abundances, and is also a measure of relative mark importance. Equivalent codes between classes of sites show differences in mark loadings. d Heatmap of standardized regression slopes for the models’ parameters, which include the four codes and motif presence (mot). Mean and standard deviation of slopes obtained from dropping each of the other covariates from the model is indicated in brackets, and is a measure of the robustness of the estimate to model specification (see Methods and Additional file 1: Methods)