Pose Filter Ensemble (PFE) plus Chemguass4 scoring method
Discovery Studio 2017/Omega 2.5/OEDocking 3.0/PFE
Omega: number of confirmations=200 (default)
OEDocking: number of docking poses=100
PFE fused with Chemgauss4 for scoring
1) All docking poses were predicted by PFE, and the native-like poses were selected and scored by Chemgauss4. Only the native-like pose of the lowest score, i.e. native pose was considered in ligand scoring (submitted)
2) The test ligands were ranked according to the Chemguass4 scores of their native poses.
No
PL/MCT-tess descriptors and libSVM based pose prediction method
Discovery Studio 2017/Omega 2.5/OEDocking 3.0/ENTESS/libSVM/Matlab
Ionization: "rule-based" method (Discovery Studio)
No tautomer was generated.
1. "Prepare ligands" module in Discovery Studio was applied to add hydrogens to all 24 ligands and generate their protonated states ("rule-based" method).
2. The construction of Pose Filter Ensemble:
1) Cocrystal structures from PDB that contained non-covalent ligands and shared the same binding site were collected. All the waters in the cocrystal structures were deleted while "clean protein" tool in Discovery Studio was applied to prepare protein structures.
2) The cognate ligand of each cocrystal structure was kept and used to define the binding site. Each cognate ligand was then redocked against the binding site of the corresponding protein structure by Omega and OEDocking.
3) Those protein structures that can generate no more than 10 binding poses were excluded, thus 8 pairs of cognate ligands and protein structures (2f1g, 2r9m, 2r9n, 2r9o, 3iej, 3kwn, 4p6e and 4p6g) were used for building pose predictors.
4) 8 Pose filters were constructed using Omega/OEDdocking/Entess/libSVM prior to their application.
5) PFE was constructed by ensembling technique that used all the predictions from 8 pose filters as inputs.
Omega: number of confirmations=200 (default)
OEDocking: number of docking poses=1000
ENTESS (descriptors that characterized protein-ligand interface: PL/MCT-tess descriptors)
Pose predictors (libSVM)
For each one of 136 test ligands,
1) Structural similarity (Tc) based on MACCS fingerprints between each test ligand and 8 cognate ligands were calculated. The cognate ligands were ranked from the most similar one to the least similar.
2) The 136 test ligands were docked to the protein structures that corresponded to their most similar cognate ligands, respectively.
3) All docking poses (100 poses per test ligand) were predicted by PFE, and the native-like poses were kept.
4) If no native pose of certain test ligands was identified from the docking poses, that test ligand was then docked to the protein structure corresponding to a less similar cognate ligand.
5) This process was not stoppted till native pose was found or all protein structures had been tested though no native pose was identified (e.g. for CatS_41).
No
Yes