Figure 2

The relationship between the area under an ROC curve (AUC; e.g. Fig. 1B) and the probability that a network classified as a mechanism is actually of that mechanism (i.e. the probability that a positive is a true positive), denoted P(True|Positive). This is calculated using Bayes’ theorem, assuming that AUC represents the accuracy of the test. The four curves show this relationship for different true frequencies of a mechanism: 50%, 10%, 1%, and 0.1%. The vertical lines show the AUC values of our classifier (Fig. 1B) for each of the five mechanisms (color), and as either an undirected or directed process (shape). Stacked lines and shapes indicate identical AUC values.