Fig. 4

The dynamics of TAD-like domains in single cells. a The cell-to-cell and cell-to-ensemble similarity of the deTOKI-predicted TAD-like domains. The single-cell data were from GM12878 [54] and compared to the ensemble in GM12878 and K562. Cell #16 was marked as it was in M/G1 phase. b The distribution of ensemble TAD boundaries over single cells. The control was set as a binomial distribution under the hypothesis that every ensemble TAD boundary has identical potential of being a single-cell TAD-like domain boundary. The vertical black lines marked 13 and 5 indicate the thresholds for the over- and under-represented boundaries in the cell population, respectively. c The distribution of bins among the classes “scSB-1,” “scSB-2,” “scSB-m,” and “absent.” The permutated control is shown in the boxplots. The distribution of H3K27me3 and H3K4me1 histone marks flanking the deTOKI-predicted single-cell-specific ensemble, scSB-m, scSB-1, and scSB-2 TAD-like domain boundaries are shown in d–g, respectively. The y-axis of panels d–g represents the mean number of peaks per bin with the same distance to the predicted TAD-like domain boundaries normalized by the average in the whole genome (MNPPB). The shadows represent 95% confidential intervals, as calculated by bootstrap. h The distribution of scSBs flanking the ensemble TAD boundaries. i The ROC curves of classification between scSB-1, scSB-2, and scSB-m based on either ChIP-seq peaks or the distance to the nearest ensemble boundaries. *P < 0.05, **P < 0.001, NS: not significant, two-sided Wilcoxon rank-sum test