Fig. 3.

A toy case showing the difference between AUC, AUPR and Coverage. The first row denotes an interaction profile between a drug and 10 targets. The second row denotes an interaction profile with one missing interaction by removing the 9-th interaction from the first row. The third row contains the predicted scores generated by performing a predicting approach on the second row. The last row lists the ranks corresponding to the predicted scores. The values of AUC, AUPR and Coverage accounting for the first row, are 0.833, 0.683 and 4 respectively. In contrast, after labelling the missing interaction as a positive correctly, those value of AUC and AUPR (corresponding to the first row) would change to 0.938, 0.912, but the value of Coverage doesn’t change. AUPR is significantly sensitive to the missing interaction, AUC is moderately sensitive and Coverage is the most robust