Mapping Chromatin Accessibility States

To facilitate comparison with MACC profiles, the signals for all external markers were re-computed for the same 300-bp bins as those used for MACC computations. The profiles around specified sets of sites were computed by using linear interpolation of MACC values associated with 300-bp bins and the resulting average profiles were additionally smoothed in the 40-bp running window. Two-state HMM and Viterbi algorithm were used to map chromatin accessibility states based on MACC profiles. The computations were performed using R-package RHmm (https://r-forge.r-project.org/projects/rhmm/). For generation of HMM based on randomized data, shuffling of MACC profile within each chromosome was used. Hierarchical clustering was performed using unweighted pair group method with the distance between profiles computed as (1—Pearson's correlation coefficients). To ensure that our findings are not biased by variability in the fragment size distributions, we reproduced crucial observations with MNase and MACC profiles computed only with fragments of mono-, di- and tri-nucleosomal lengths (Supplementary Figs 21–25).