The observed cumulative fractions of cell-type specificity are a function of the composition of cell-types in the compendium and do depend to some extent on the variability of data quality for the different marks. For example, the enhancer mark (H3K4me1) does have a much better signal-to-noise ratio than the transcribed mark (H3K36me3). One might expect this to result in more spurious variation of states associated with the transcribed mark. However, contrary to this expectation, the cumulative fractions for states involving only the transcription mark (Tx and TxWk) and not the enhancer mark indicate that these states are in fact less variable and more constitutive across cell types. On the other hand, all states composed of the enhancer mark (H3K4me1), irrespective of whether they do (TxFlnk, EnhG) or do not (EnhBiv, Enh, BivFlnk, TssAFlnk) include the transcription mark (H3K36me3), are far more cell-type specific. These observation indicate that the increased variability of states are largely due to the enhancer mark (H3K4me1) than the transcribed mark (H3K36me3). As replicates are not available in all epigenomes, we did not correct for inter-replicate variation in this analysis, but in the state switching analysis below, we utilize samples from the same tissue as quasi-replicates.
