Comparison of the Predictions at Different Levels of Confidence

METHODS The four sets of predicted CTFPs analyzed in this paper were identified at a level of confidence deemed as appropiated by the authors of each method. These predictions, used in the main text of our paper, will hereinafter be referred to as main sets. However, predictions at other levels of confidence (higher or lower, or both) were supplied as supplementary data accompanying the papers under study. Method N provided a more stringent set of predictions by integrating protein location data and protein function data to the predictive conditions. This set contained 39 CTFPs. For short, we will refer to this set as integrated set N. Predictions made by method B were dependent in a p-value (chosen to be PB<0.001 in the paper). The authors provided a more stringent list of predictions (at PB<0.0001) and a less stringent list of predictions (at PB<0.01). The low-confidence set B contained 208 CTFPs and the highconfidence set B contained 16 CTFPs. As noted by the authors, the functional nature of the TF pairs included in each set was different. Authors of method T provided as supplementary data a list of doubtful predictions which consisted of 29 CTFPs. This set will be called doubtful set T. Finally, authors of method C did not explicitly provide predictions at different levels of confidence. They did, however, classify some of their predictions as false positives on the basis of literature support. Hence, we have removed those false positives from their main set of predictions. This set resulted in a total of 44 CTFPs. Because of its nature, this set will be referred to as true positive set T. Overlap between the sets was calculated by means of the Jaccard coefficient (see Methods in the main text).