MMS DockBench CrossDocking Method 1


Maestro 2016/OpenEye 1.9.2/DockBench 1.01 [1]/Autodock 6.5/GOLD 5.4.1/MOE 2015.1001/Plants 1.2/RDock 2013.1

System Preparation Parameters

Assumed pH 7.4

System Preparation Method

Ligands were prepared using LigPrep Tool of Maestro, retaining specified chiralities and without tautomers generations. Strong acids were deprotonated and strong bases protonated using Wash tool of MOE. Ligands were minimized with MMFF94 force field.
Proteins were prepared by selecting chain A of crystallographic structures and by using Structure Preparation Tool of MOE. Protonation states were assigned with Protonate 3D tool of MOE.

Pose Prediction Parameters

DockBench Num_poses= 20
DockBench Radius= 20
DockBench proteins pdb (21 pdbs) = 3bej,3dct,3dcu,3fli,3fxv,3dg2,3hc5,3hc6,3oki,3olf,3omk,3omm,3oof,3ook,3p88,3p89,3rut,3ruu,3rvf,4qe6,4qe8
DockBench docking protocols= autodock-ga,autodock-lga,autodock-ls,glide-sp,gold-asp,gold-chemscore,gold-goldscore,gold-plp,moe-affinitydg,moe-gbviwsa,moe-londondg,plants-chemplp,plants-plp,plants-plp95,rdock-solv,rdock-std
Key parameters of the docking protocols are specified in DockBench manuscript [1].
CrossDocker [DockBench] = same as above

Pose Prediction Method

Crystallographic structures of FXR were retrived from the PDB and used for a docking benchmark. After protein superposition of the pdb complexes, the co-crystalized ligands of the pdb structures were grouped into clusters on the basis of their shape similarity. RDKit ShapeTanimotoDistance was computed among the ligands. The resulting distance matrix was analyzed with scikit-learn DBSCAN clustering algorithm, using a cutoff value of 0.45: 6 clusters were obtained (RESULTS-Clusters).
A cross docking benchmark was applied, in order to identify which was the "protein-docking protocol" couple that was able to better reproduce each co-crystalized conformation.
The evaluation of the docking protocols performances in reproducing the crystallographic ligand conformation was performed by computing the RMSD of the docking poses to the references. More in details for each pdb structure, only the ligands that were members of the same cluster of the co-crystalized ligand were considered. Thus for each member of the cluster, the RMSD of the top five poses were computed, and their average was used as score.
In the cross docking approach the original ligand was excluded from the ligands subset considered in the calculation of the final RMSD. For this reason, for the ligands that were part of one member-only cluster, self-docking results were used.
In addition we computed an other evaluation, defined here as "total-cross-docking", in which for the RMSD calculation, we considered all the ligands despite their cluster membership.

The best "protein-docking protocol" couple (winner) was identified as minimum RMSD average for each cluster (RESULTS-Clusters). Consequently ligands FXR1 to FXR36 were linked to each cluster on the basis of the ROCS TanimotoComboShapeSimilarity to the winner. A cutoff value of 0.8 for TanimotoComboShapeSimilarity was used to achieve the cluster population.

For each FXR ligand there were three possible results for the previous step:
1) ligand linked to a cluster
2) ligand linked to a one-member cluster
3) ligand orphan of a cluster

In case 1) the "protein-docking protocol" winner for the cluster was selected; in case 2) the self docking results were considered; in case 3) the best "protein-docking protocol" in "total-crossing-docking" was considered

For FXR1 to FX36 ligands the docking simulations were performed by using the "protein-docking protocol" suggested by our analysis (RESULTS-FX1to36). The top five poses were selected for the submission.


Cluster 1: 3RVF, 3DCT, 3HC6, 3RUU, 3RUT, 3FXV, 3GD2, 3HC5, 3P89, 3DCU, 3P88
Cluster 2: 4QE8
Cluster 3: 3FLI
Cluster 4: 4QE6
Cluster 5: 3BEJ
Cluster 6: 3OKI, 3OLF, 3OMK, 3OMM, 3OOF, 3OOK

Protein Docking Protocol reference Result
Cluster 1: 3gd2 gold-plp (cross docking)
Cluster 2: 4qe8 plants-plp95 (self docking)
Cluster 3: 3fli rdock-solv (self docking)
Cluster 4: 4qe6 gold-goldscore (self docking)
Cluster 5: 3bej gold-asp (self docking)
Cluster 6: 3omk gold-plp (cross docking)

Outliers: 3gd2 gold-goldscore (total-cross-docking)

Ligand Protein Docking Protocol
FX1 3gd2 gold-goldscore
FX2 3omk gold-plp
FX3 3fli rdock-solv
FX4 3omk gold-plp
FX5 3fli rdock-solv
FX6 3omk gold-plp
FX7 3omk gold-plp
FX8 3omk gold-plp
FX9 3omk gold-plp
FX10 3gd2 gold-goldscore
FX11 3gd2 gold-goldscore
FX12 3gd2 gold-goldscore
FX13 3omk gold-plp
FX14 3omk gold-plp
FX15 3gd2 gold-goldscore
FX16 3bej gold-asp
FX17 3omk gold-plp
FX18 3omk gold-plp
FX19 3omk gold-plp
FX20 3omk gold-plp
FX21 3omk gold-plp
FX22 3omk gold-plp
FX23 4qe8 plants-plp95
FX24 3omk gold-plp
FX25 3omk gold-plp
FX26 3omk gold-plp
FX27 3omk gold-plp
FX28 3omk gold-plp
FX29 3omk gold-plp
FX30 3omk gold-plp
FX31 3omk gold-plp
FX32 3omk gold-plp
FX33 3gd2 gold-plp
FX34 4qe6 gold-goldscore
FX35 3omk gold-plp
FX36 3omk gold-plp

[1] Cuzzolin, A.; Sturlese, M.; Malvacio I.; Ciancetta, A.; Moro S. DockBench: An integrated Informatic Platform Bridging the Gap between the Robust Validation of Docking Protocols and Virtual Screening Simulations. Molecules 2015, 20 (6), 9977-9993.