To avoid spurious switching due to differences in data quality, we restricted this analysis to chromatin states from the 43 high quality and non-redundant Roadmap epigenomes (see above). Using the 15 state primary model, we computed the empirical switching frequency of any pair of states across all pairs of 43 epigenomes. For a given pair of states A and B, we counted the number of genomic bins that were labeled as (A,B) or (B,A) in all pairs of epigenomes. The switching frequency matrix (which is symmetric) was then row-normalized in order to convert the switching frequencies to switching probabilities. This is done to avoid a dependence on the total number of epigenomes. Also, the switching probabilities unlike switching frequencies are not dominated by states that are highly prevalent (e.g. quiescent state). Fig. S7b. shows the empirical switching probabilities for all pairs of states across the 43 epigenomes. In order to differentiate between chromatin state dynamics across tissues (inter-tissue) relative to variation of states across individuals or replicates from the same tissue (intra-tissue), we also computed analogous switching frequencies by restricting
