Gypsum-DL is a free, open-source program for preparing 3D small-molecule
models. Beyond simply assigning atomic coordinates, Gypsum-DL accounts for
alternate ionization, tautomeric, chiral, cis/trans isomeric, and
ring-conformational forms. It is released under the Apache License, Version
2.0 (see LICENSE.txt
).
If you use Gypsum-DL in your research, please cite:
Ropp, Patrick J., Jacob O. Spiegel, Jennifer L. Walker, Harrison Green, Guillermo A. Morales, Katherine A. Milliken, John J. Ringe, and Jacob D. Durrant. (2019) "Gypsum-DL: An Open-source Program for Preparing Small-molecule Libraries for Structure-based Virtual Screening." Journal of Cheminformatics 11:1. doi:10.1186/s13321-019-0358-3.
Ropp PJ, Kaminsky JC, Yablonski S, Durrant JD (2019) Dimorphite-DL: An open-source program for enumerating the ionization states of drug-like small molecules. J Cheminform 11:14. doi:10.1186/s13321-019-0336-9.
To run Gypsum-DL, acquire a copy of this repository, either by git clone or by download. Install the required dependencies via your favorite python package manager. We suggest using Anaconda to manage packages:
conda install -c rdkit rdkit numpy scipy mpi4py
If you encounter an error like ImportError: libboost_python3.so.1.65.1: cannot open shared object file: No such file or directory
, a helpful
user
suggested installing RDKit this way instead:
conda create -c conda-forge --name gypsum_dl_env rdkit numpy scipy mpi4py -y
conda activate gypsum_dl_env
Gypsum-DL accepts the following command-line parameters:
-h, --help show this help message and exit
--json param.json, -j param.json
Name of a json file containing all parameters.
Overrides all other arguments specified at the
commandline.
--source input.smi, -s input.smi
Name of the source file (e.g., input.smi). Note:
support for SMI (SMILES) files is better than support
for SDF files, though Gypsum-DL can handle both.
--output_folder OUTPUT_FOLDER, -o OUTPUT_FOLDER
The path to an existing folder where the Gypsum-DL
output file(s) will be saved.
--job_manager {mpi,multiprocessing,serial}
Determine what style of multiprocessing to use: mpi,
multiprocessing, or serial. Serial will override the
num_processors flag, forcing it to be one. MPI mode
requires mpi4py 2.1.0 or higher and should be executed
as: mpirun -n $NTASKS python -m mpi4py
run_gypsum_dl.py ...-settings...
--num_processors N, -p N
Number of processors to use for parallel calculations.
--max_variants_per_compound V, -m V
The maximum number of variants to create per input
molecule.
--thoroughness THOROUGHNESS, -t THOROUGHNESS
How widely to search for low-energy conformers. Larger
values increase run times but can produce better
results.
--separate_output_files
Indicates that the outputs should be split between
files. If true, each output .sdf file will correspond
to a single input file, but different 3D conformers
will still be stored in the same file.
--add_pdb_output Indicates that the outputs should also be written in
the .pdb format. Creates one PDB file for each
molecular variant.
--add_html_output Indicates that the outputs should also be written in
the .html format, for debugging. Attempts to open a
browser for viewing.
--min_ph MIN Minimum pH to consider.
--max_ph MAX Maximum pH to consider.
--pka_precision D Size of pH substructure ranges. See Dimorphite-DL
publication for details.
--skip_optimize_geometry
Skips the optimization step.
--skip_alternate_ring_conformations
Skips the non-aromatic ring-conformation generation
step.
--skip_adding_hydrogen
Skips the ionization step.
--skip_making_tautomers
Skips tautomer-generation step.
--skip_enumerate_chiral_mol
Skips the ennumeration of unspecified chiral centers.
--skip_enumerate_double_bonds
Skips the ennumeration of double bonds.
--let_tautomers_change_chirality
Allow tautomers that change the total number of chiral
centers (see README.md for further explanation).
--use_durrant_lab_filters
Use substructure filters to remove molecular variants
that, though technically possible, were judged
improbable by members of the Durrant lab. See
README.md for more details.
--2d_output_only Skips the generate-3D-models step.
--cache_prerun, -c Run this before running Gypsum-DL in mpi mode.
--test Tests Gypsum-DL to check for programming bugs.
Prepare a virtual library and save all 3D models to a single SDF file in the present directory:
python run_gypsum_dl.py --source ./examples/sample_molecules.smi
Instead save all 3D models to a different, existing folder:
python run_gypsum_dl.py --source ./examples/sample_molecules.smi \
--output_folder /my/folder/
Additionally save the models associated with each input molecule to separate files:
python run_gypsum_dl.py --source ./examples/sample_molecules.smi \
--output_folder /my/folder/ --separate_output_files
In addition to saving a 3D SDF file, also save 3D PDB files and an HTML file with 2D structures (for debugging).
python run_gypsum_dl.py --source ./examples/sample_molecules.smi \
--output_folder /my/folder/ --add_pdb_output --add_html_output
Save at most two variants per input molecule:
python run_gypsum_dl.py --source ./examples/sample_molecules.smi \
--output_folder /my/folder/ --max_variants_per_compound 2
Control how Gypsum-DL ionizes the input molecules:
python run_gypsum_dl.py --source ./examples/sample_molecules.smi \
--output_folder /my/folder/ --min_ph 12 --max_ph 14 --pka_precision 1
Run Gypsum-DL in serial mode (using only one processor):
python run_gypsum_dl.py --source ./examples/sample_molecules.smi \
--job_manager serial
Run Gypsum-DL in multiprocessing mode, using 4 processors:
python run_gypsum_dl.py --source ./examples/sample_molecules.smi \
--job_manager multiprocessing --num_processors 4
Run Gypsum-DL in mpi mode using all available processors:
mpirun -n $NTASKS python -m mpi4py run_gypsum_dl.py --source ./examples/sample_molecules.smi \
--job_manager mpi --num_processors -1
Gypsum-DL can also take parameters from a JSON file:
python run_gypsum_dl.py --json myparams.json
Where myparams.json
might look like:
{
"source": "./examples/sample_molecules.smi",
"separate_output_files": true,
"job_manager": "multiprocessing",
"output_folder": "/my/folder/",
"add_pdb_output": true,
"add_html_output": true,
"num_processors": -1
}
Gypsum-DL is designed to process drug-like molecules. Generating 3D structures
for larger molecules takes a very long time. For example, in our tests it
takes Gypsum-DL a very long time to process this molecule:
CCCC[C@@H](C(N[C@H]1CC(NCCCC[C@H](NC([C@@H](NC([C@@H](NC([C@@H](NC([C@@H]2CCCN2C1=O)=O)Cc3ccccc3)=O)CCCNC(N)=N)=O)Cc(c[nH]4)c5c4cccc5)=O)C(N6CCC[C@H]6C(N[C@@H](C(C)C)C(N)=O)=O)=O)=O)=O)NC(C)=O
You may wish to run your compounds through a drug-like filter before processing them with Gypsum-DL.
Gypsum-DL uses MolVS to generate tautomers. While MolVS is effective, we have
noticed that it sometimes generates inappropriate tautomers that change the
total number of chiral centers, e.g. O=C(c1ccc(CN)cc1)N
to
N=Cc1ccc(C(O)N)cc1
. But some legitimate tautomers also change the number of
chiral centers, e.g., C[C@@H](C(C)=O)F
to C/C(F)=C(C)\O
.
To compensate for this MolVS bug, by default Gypsum-DL rejects all tautomers
that change the total number of chiral centers. Use the
--let_tautomers_change_chirality
flag if you would like to retain these
tautomers instead. As always, be sure to examine the structures that Gypsum-DL
outputs to ensure they are chemically feasible.
In looking over many Gypsum-DL-generated variants, we have identified a number of substructures that, though technically possible, strike us as improbable or otherwise poorly suited for virtual screening. Here are some examples:
C=[N-]
[N-]C=[N+]
[nH+]c[n-]
[#7+]~[#7+]
[#7-]~[#7-]
[!#7]~[#7+]~[#7-]~[!#7]
[#5]
(boron)O=[PH](=O)([#8])([#8])
N=c1cc[#7]c[#7]1
[$([NX2H1]),$([NX3H2])]=C[$([OH]),$([O-])]
- Metals
If you'd like to discard molecular variants with substructures such as these,
use the --use_durrant_lab_filters
flag.
Some users have reported that Gypsum-DL fails to produce 3D models when
processing molecules with highly constrained ring systems, such as amantadine
compounds (e.g., this molecule from ChemBridge:
CC1=CC=CN2N=CC(C(=O)NC34CC5CC(C3)CC(C5)(C4)N3C=NC=N3)=C12
). Increasing the
thoroughness
parameter may help in these cases.
When processing large libraries, Gypsum-DL requires substantial memory. Some users have reported that the program suddenly stops in these situations. To correct the problem, either increase the available memory, or divide your library into several smaller files and processes them sequentially.
Gypsum-DL aims to enumerate many possible variant forms, including forms that are not necessarily probable. Beyond applying Durrant-Lab filters, several methods allow users to exclude other potentially problematic forms:
- Identify the steps Gypsum-DL takes to generate a given problematic form
(see the "Genealogy" field of every output SDF file). Then use parameters
such as
--skip_optimize_geometry
,--skip_alternate_ring_conformations
,--skip_adding_hydrogen
,--skip_making_tautomers
,--skip_enumerate_chiral_mol
, or--skip_enumerate_double_bonds
to skip the problem-causing step. This fix is easy, but it may unexpectedly impact unrelated compounds. - Consider adjusting the
--min_ph
,--max_ph
, or--pka_precision
parameters if Gypsum-DL is producing compounds with undesired protonation states. Alternatively, you can delete specific protonation rules by modifying thegypsum_dl/Steps/SMILES/dimorphite_dl/site_substructures.smarts
file. - Add to the Durrant-Lab filters if there is a specific substructure you
would like to avoid (e.g., imidic acid due to amide/imidic-acid
tautomerization). Simplify modify the
gypsum_dl/Steps/SMILES/DurrantLabFilter.py
file.