Figure 4.

Chromatin factors associated with open chromatin and enhancers positively correlate with condensin binding. (A) The number of condensin II peaks are plotted with respect to the linear length of each chromosome. (B) The number of condensin II peaks are plotted against the length of the chromosome that is covered by H4K8Ac and H4K16Ac peaks. A trend line fitted to autosomal data indicates a positive correlation (R2 = 0.9). (C) Condensin binding within 1 kb contiguous windows across the X chromosome (left) and autosomes (right) positively correlate with active marks (for example, H3K27ac, H2A.Z) and negatively correlate with repressive marks (for example, H3K27me3, H3K9me3). (D) An ensemble classifier (random forests) was learned to predict condensin binding across the genome. The top 20 features (among 92 total features, Additional file 1: Table S1) with the highest predictive power are plotted for condensin I-IDC (left) and condensin II (right) with the most important feature on top. The features are ranked based on the mean decrease in accuracy, which describes the difference between the error-rate of the actual classification and the error-rate after permuting the feature, averaged over all classifiers (trees).

Kranz et al. Genome Biology 2013 14:R112   doi:10.1186/gb-2013-14-10-r112
Download authors' original image