Classifying stable nucleosomes containing H2A/H4R3me2s or H2A.Z using histone modification features. (a) Receiver Operating Characteristic (ROC) curve, demonstrating classifier performance. (b) Random forests feature importance by mean decrease in Gini index. Features have a higher frequency in H2A/H4R3me2s nucleosomes (red) or H2A.Z nucleosomes (blue). The dashed, vertical line shows the estimated (permutation-based) significance threshold after multiple testing correction. (c) A classification tree with splits (no borders) and leaves (borders), below which is the number of nucleosomes classified correctly and, in parentheses, incorrectly at that stage. Leaves show the predicted class labels of nucleosomes partitioned there. Splits show the condition that best separates the data. Branch labels indicate the directions in which the split condition is true (“yes”) and false (“no”).
Capurso et al. BMC Genomics 2012 13:630 doi:10.1186/1471-2164-13-630