Fig. 5

Overview of the workflow presented in this manuscript. In the data preparation step, we randomly compile disjoint training data and testing data each with positive alignments and negative alignments for each of the transcription factors CTCF, GABP, NRSF, SRF, and STAT1. In the model training step, we train each of the four presented foreground models as well as a background model by expectation maximization with 150 restarts. We choose the foreground model and the background model with maximum likelihood, classify the testing data using a likelihood-ratio classifier, and extract different characteristics such as the ROC curve, the PR curve, the inverse temperature, and the inferred motif. We repeat the described procedure 100 times and calculate mean values and standard errors for several quantities such as the areas under the ROC curves or the PR curves