A hierarchical docking method: XDZ_2
OpenEye/MGLTools/Smina/Tensorflow
Gasteiger charges
All the released crystal structures of human BACE1
protein-small molecule complexes were collected from the Protein Data Bank.
Preparation for docking with Smina was done using MGLTools.
Exhaustiveness=50 #exhaustiveness of global search (default=8)
BACE1_CNNMode model + BACE1_CNNScore model
Num_modes=100 #max number of poses to generate
Energy_range=10 #energy difference (kcal/mol) between the best and worst binding mode
For a query ligand, the OEChem TK in OpenEye was employed to calcualte ligand similarities
to the small molecules in the released BACE1 structures.
The PDB entry that has the best similarity score with the query ligand was used as the receptor structure for Smina docking.
The binding modes were classified into near-native and non-near-native binding modes with BACE1_CNNMode model,
a convolutional neural networks (CNN) model for pose prediction that was developed using the released BACE1-ligand structures.
Then, the near-native models were ranked with the BACE1_CNNScore model,
a CNN-based scoring function that was developed using the released BACE1-ligand structures and their corresponding binding affinities.
No
Yes