Fig. 10.

AUPR scores obtained with a 5-fold nested cross-validation scheme for the $S_1$ (random split), $S_2$ (orphan protein in test set), $S_3$ (orphan molecule in test set), and $S_4$ (double orphan) settings on the DBEColi dataset. The performance of $CN\text{-}currPCBA$ and $CN\text{-}currDBHuman$ are compared to the reference shallow methods (kronSVM and NRLMF) and to FNN and CN as reference methods for deep learning. For each setting, the order from left to right in which the results are displayed is indicated in legend