Score metrics



The decision which probability to use as realistic cutoff is far from trivial. It depends to a large degree on the preferences and demands of the user, since any cutoff represents a compromise between true and false positives. To assist the user making this decision we make the dependence between the score and the ratio TP/FP transparent.

Below you can see for both methods a colored ROC plot for overview, a download for detailed consideration of the underlying data and finetuning of the cutoff and finally a comparison of different statistical quality measures. In the colored ROC plots, the color scale represents the choice of cutoff. The plots wer created using the prettyPROC R-package.


Random Forest






Download ROC data for RF.






Graph neural networks





Download ROC data for GNN.





Please cite: Sebastian Canzler, Markus Fischer, David Ulbrecht, Nikola Ristic, Peter W Hildebrand, René Staritzbichler: ProteinPrompt: a webserver for predicting protein-protein interactions, bioRxiv 2021.09.03.458859