Figure 2.

Feature importance computed from Boruta package. Relative importance of different features as computed by the Boruta package [29]. Each boxplot corresponds to a different feature and represents importance z-score from 500 randomizations. Histone modifications are the most important (z-score above 10), followed by all motif features (z-score above 3), all of which are separated from the randomized control variables with (red, z-scores below 3).