Classification results on the H1-hESC test data set.

We display the sensitivity (for specificity of 99%) on the independent test data set . Here we pick the optimal model from the cross validation experiment (green) and compare it with the PWM model (blue). We observe the PMM yielding a more than 10% higher sensitivity than the PWM model, which shows that the PMM picked via cross validation on improves motif discovery compared to the PWM model. In addition, we display the sensitivity of alternative models that take into account statistical dependencies. We observe that alternative models also benefit from taking into account statistical dependencies, even though to a lesser extent than a PMM.