Figure 4.

Comparison between $DCC$ values achieved by DeepSite and our model: (A) on the whole test set and (B) zoomed-in on high-quality predictions. Each point corresponds to $DCC$ values obtained with the two models. Binding sites for which U-Net did not return predictions are depicted with red crosses at the top of the image. Structures that were used to train DeepSite are colored with orange. Gray area marks 5 Å difference in both directions from the diagonal (same predictions for the two models). Predictions for L-histidinol dehydrogenase and 7,8-dihydroneopterin aldolase are marked with ellipses and annotated with proteins’ short names (HDH and DHNA, respectively).