Ensemble-docking with AutoDock4.2 and enhanced-sampling of Pocket Shape and further relaxation of poses by molecular mechanics
VMD 1.9.3
Pymol 1.8.4
MGLTools 1.5.6
Amber18
Gromacs2016.5
R3.5.1
PLUMED2.4
AutoDock4.2
OpenEye Omega 2018.Feb.1
OpenEye Shape 2018.Feb.1
OpenEye ROCS 3.2.2.2
ChemmineR 3.32.1
fmcsR 1.22
Assumed pH neutral
Protein charges from AMBER-FB15 force field for MD simulations and Gasteiger for docking
Ligand parameters and charges (Gasteiger) derived using the MGLTools
Ligand parameters and charges (bcc) for docking pose optimization were derived using the antechamber tools
The sequence given in "BACE_target_D3R_GC4.fasta" file was used to search for
structural templates in the pdb databank, using a threshold of 95% for the sequence identity and requiring the presence
of the word "BACE" in the title. The structure with PDB ID 1SGZ, free of any ligand, was thus selected as structural template.
Molecular dynamics simulations of the protein: standard MD simulations of 1 microsecond in length were performed with the pmemd
program of the amber18 suite, using a 0.15 M KCl water solution. The AMBER-FB15 force field with added atom types for amino
acid side chains was used for the protein, together with the TIP3P-FB water model and ions.
In addition, enhanced-sampling MD simulations (namely bias-exchange/well-tempered metadynamics) were performed using GROMACS2016.5
and the PLUMED2.4 plugin to enhance sampling of the putative binding site.
The latter was identified as the union of the binding pockets (residues within 3 Angstrom from the ligand) in experimental structures
with PDB IDs: 2IQG, 3DV1, 3DV5, 3K5C, 3VEU, 4DPI, 4KE1, 6BFD. These in turn were identified using the advanced search functionality
of the RCSB PDB and the following settings: 95% sequence identity and presence of a ligand, which gave 340 templates.
A single receptor was then selected for each target ligand, based on the similarity of the crystallographic ligand to the competition
ligand as calculated by the MCS tanimoto index between every template ligand and the full set of 340 complexes.
The list of residues was then manually adjusted to keep the number around 20. Namely, the residues list has entries:
12 13 32 34 35 71 72 73 74 108 109 110 198 226 228 229 230 233 329 332.
The BACE ligands were converted from SMILES format to 3D structures using OE-Omega and sampling up to 500 conformers per ligand.
We selected 10 poses for docking based on the tanimoto-combo (shape + colour) similarity between the generated ligand poses and the
selected (see above) template ligand. Each PDB file was converted to PDBQTs by assigning charges and atom types, and the box size was
determined to cover all the residues lining the binding site.
Protein conformations were generated from a cluster analysis on the binding site residues from the cumulative trajectory obtained by
concatenating the standard and metadynamics ones. The distance RMSD (dRMSD) matrix was used as metric to cluster the conformations,
and imposing a total of 200 clusters to be produced. The pdb files were converted to PDBQT files using the MGLTools.
Hybrid genetic algorithm with local search (GA-LS)
ga_num_evals 25000000 (default 2500000)
autodock_parameter_version 4.2 (AD4.1_bound.dat)
ga_run=1 # do this many hybrid GA-LS runs
grid spacing 0.25 Angstrom (default 0.375)
Ensemble-docking calculations were performed with AutoDock4.2, disallowing any flexibility of both
ligands and receptors. Docking runs were executed with the above specified parameters while default values were applied for the rest
of the variables. For each compound, 10 conformations and 200 receptor structures were employed in ensemble-docking runs.
The top 5 clusters of poses from the docking (see above) are submitted with this protocol. After structural alignment of the binding sites
of all generated complexes, the poses were clustered using a distance RMSD cutoff of 0.075 times the number Nha of heavy atoms of the
ligand, in order to tune the cutoff the the size of the compound. Clusters were ordered according to the top score (lowest binding free energy)
within each cluster.
The top 10 poses were optimized through three cycles of restrained optimization (restraints of 0.3, 0.2 and 0.1 kcal/mol
on all heavy atoms of protein and ligand in the three steps, respectively), and a unrestrained optimization. Poses were then rescored with
AutoDock4.2
No
Yes