1147-1-my4m5-PosePredictionProtocol.txt

Name

Crystal Alignment

Software

Flare 2.0 revision 34140
fkcombu
RDkit 2018.09.01

System Preparation Parameters

Assumed pH 4.5

System Preparation Method

Ligands were read from smiles files with Flare, protonated to a pH of 4.5 and minimised with Flare. They were then docked to, 2HHN with only retaining one docking pose. The resulting ligand positions were saved.

Pose Prediction Parameters

default of fkcombu

Pose Prediction Method

The preliminary docked ligands were used to find the maximum common substructure between ligand and available crystal structures of GC3. Ligands of GC4 were matched to GC3 crystals based on the largest MCS. Matched GC4 ligands were aligned to their crystal structure match using the crystal structure receptor as the reference protein.

Answer 1

no

1147-4-5otd6-FreeEnergyProtocol.txt

Name

Alchemical free energies using BioSimSpace.

Software

BioSimSpace (feature-freenrgy) https://github.com/michellab/BioSimSpace/tree/feature-freenrg
Sire (feature-mapping) https://github.com/michellab/Sire/tree/feature-mapping
AmberTools (18)
GROMACS (2018)
pymbar (3.0.3)
freenrgworkflows (1.1)
FESetup 1.2.1, SUI version: 0.8.3
RDkit 18.09.01

Parameter

timestep 2.00 femtosecond
temperature = 300.00 kelvin
pressure = 1 atm
reaction field dielectric = 78.3
cutoff type = cutoffperiodic
cutoff distance = 10 angstrom
minimise = True
lambda values between 9 and 17 windows.

Method

Full installation instructions:
https://github.com/michellab/D3R2018/blob/master/CatS/BSS/README.md

The ligands were prepared for relative free energy calculations from the output
of the docking protocols. Each ligand was parameterised using GAFF within
BioSimSpace using the following script:
https://github.com/michellab/D3R2018/blob/master/CatS/BSS/parameterise.py
This was run for each of the 39 ligands as well as two additional intermediate
states that were added in order to simplifly perturbations between certain
ligand pairs.

Full production simulations for each forward and reverse ligand pair mapping
within the network were run almost entirely within BioSimSpace using the
following script:
https://github.com/michellab/D3R2018/blob/master/CatS/BSS/binding_freenrgy.py
This performed the following sequence of tasks:

1) Read in the protein and crystal waters from a PDB file. Extract the
water molecules from the system.
2) Parameterise the protein using AMBER FF14SB.
3) Read the two pre-parameterised ligands. (GAFF2)
4) Look for a mapping file specifying the matching atoms in the maximum
common substructure of the ligand pair. If the file is not present,
then the mapping is generated by BioSimSpace. Preliminary simulations
found low atom counts for certain BioSimSpace generated mappings so,
where necessary, mapping files were generated using additional tools
(FESetup and RDKit). However, it should be noted that all tools struggled
with certain mappings, with each performing better or worse for
different pairs.
5) The first ligand is aligned to the second based on the mapping.
6) The two ligands are merged based on the mapping. This creates a
perturbable molecule describing the two lambda end states (0 and 1)
as well as the properties, e.g. bonds, angles, dihedrals, etc., that
are perturbed.
7) A molecular system is created from the merged molecule, the protein,
and the crystal waters.
8) This system is then solvated in a 60 Angstrom cubed box of TIP3P
water. Existing crystal waters are automatically converted to the
correct water topology and the system is neutralised.
9) A free energy protocol is created using a run time of 4 nanoseconds
per window and 17 linearly spaced windows per free energy leg.
10) Using the solvated protein-ligand complex and protocol, BioSimSpace
automatically configures all of the input required for the full
binding free energy calculation.
11) The simulation is run. For simplicity, each lambda window is run in
sequence. BioSimSpace can automatically detect crashes and re-run
specific windows as necessary.

The full set of simulations were run on a 10 node GPU cluster with 2
GPUs per node. This ran on the Oracle Cloud infrastructure and was set up
using the following guide:
https://cluster-in-the-cloud.readthedocs.io/en/latest/running.html#slurm-jobs
A single job submission involved running the full forward
and reverse mappings on the two GPUs within a node. A Slurm template and
the bash script used for submission can be found at the following links:
https://github.com/michellab/D3R2018/blob/master/CatS/BSS/template.slm
https://github.com/michellab/D3R2018/blob/master/CatS/BSS/submit.sh

The output the simulations was analysed using the 'analyse_freenrg'
package within Sire, which uses the pymbar package to perform multi
state Bennet's acceptance ratio analysis. The following bash script
was used to perform incremental analysis as jobs finished:
https://github.com/michellab/D3R2018/blob/master/CatS/BSS/analyse_data.sh

Answer 1

No

Answer 2

No