Question: About the evaluation protein selection #6
Description
Dear authors, thank you for your contributions.
I would like to ask a question about the selection of evaluation proteins in your work.
I have read the dedicated part in the appendix:
Protein targets. Three protein targets fa7 (FA7), parp1 (PARP-1), 5ht1b (5-HT1B) are chosen as design objectives to train the generative model on. While the model performance can greatly deviate according to its protein target, we carefully chose the targets to avoid bias in the experiments. The three targets have one of the highest AUROC scores when the protein-ligand binding affinities for DUDE+ ligands are approximated with AutoDock Vina and the result was compared with ground truth, meaning that AutoDock Vina works fairly well for those three targets [57]. We assume that QuickVina 2, a derivative of AutoDock Vina, would similarly work well for the three targets. Also, three targets have different protein family memberships - fa7 (Coagulation factor VII) in protease family, parp1 (Poly [ADP-ribose] polymerase-1) in polymerase family, and 5ht1b (5-hydroxytryptamine receptor 1B) in G protein-coupled receptor family.
The reference 57 here is the DUD-E+ paper. While I understand the method for choosing the protein. I don't see any results in the paper on AutoDock Vina. The results are on DOCK, Glide and Surfex.
Can you provide some information on how you got the results on AutoDock Vina?