ICM/APF 3D QSAR + MMGBSA
Molsoft ICM 3.8-5
default
MM/GBSA model was trained on the available FXR activity data in ChEMBL database. Training set was assembled using only ligands within 0.2 Tanimoto distance from any of the 102 query compounds. Resulting training set consisted of 115 ligands. Training set ligands were docked to a set of FXR pockets as desribed in PosePredictionProtocol.txt. MM/GBSA type score was trained by PLS regression on three terms: van der Waals interaction, buried hydrophobic surface and generalized Born solvation electrostatics. Regression was used to perform prediction for 102 query compounds on the poses generated as described in PosePredictionProtocol.txt.
ICM Dock
Molsoft ICM 3.8-5
default
Standard ICM conversion procedure was used to add hydrogens, assign atom types and charges
All the waters were removed.
Monte Carlo search 'effort' parameter was set to 10.
Docking runs were executed with the above specified parameters while default values
were applied for the rest of the variables. Each ligand was docked to a set of receptor conformations from X-ray structures in PDB and structures released by D3R after Stage 1.
Only poses with APF (Atomic Property Fields) similarity >0.4 to the cognate ligand of the corresponding receptor conformation were retained.
The top 5 poses according to ICM VLS score were further analysed: poses were re-ranked according to composite ICM-VLS/APF score (the product of physics-based ICM VLS Score and APF similarity to the cognate ligand of the respective receptor conformation). Poses were also clustered across the compound set to identify consensus poses for each chemotype (when applicable). In a few cases where best-scoring pose disagreed with the chemotype consensus, the pose in agreement with the chemotype was selected instead.