For each state s for the core 15 state joint model, we computed the number of genomic bins that were labeled with that state in at least one epigenome (Gs). From amongst these bins we counted the number of bins (gs,i) that were labeled as being in state s in exactly i epigenomes (i=1..127). We converted these counts to fractions (gs,i / Gs) and computed the cumulative fraction that is consistently labeled with the same chromatin state in at most N epigenomes (N=1..127). States whose cumulative fractions rise faster than others represent those that are less constitutive (more variable). We repeated the same procedure restricted to 43 high quality and non-redundant Roadmap epigenomes (using only 1 representative epigenome from those corresponding to the ESC lines, iPS lines or epigenomes for the same tissue type from different individuals and excluding ENCODE cell-lines) (Table S1 - Sheet VariationAnalysis) (Fig. S6a). Analogous analysis was performed on states from the 18 state expanded model (Extended Data 5a, Fig. S6b).
