Format of a predicted network in the in silico network inference challenge.

Predicted edges were to be ranked from most confidence to least confidence that the edge is present in the network. A directed edge is denoted by a source and target node and an arbitrary (non-increasing) score between one (most confidence) to zero (least confidence). Thus, edges that are predicted to exist in the network should be at the top of the list and those predicted not to exist in the network should be at the bottom of the list. To evaluate the predicted network, two metrics—area under the ROC curve and area under the precision-recall curve—were computed by scanning all possible decision boundaries (i.e., k = 1, k = 2, etc.) up to the maximum number of possible directed edges (excluding self-edges).