Software pipeline used to generate diverse protein structural ensembles, for the purpose of seeding multiple parallel molecular dynamics simulations, and subsequent construction of Markov state models.
- Daniel L. Parton | [email protected]
- John D. Chodera | [email protected]
- Patrick B. Grinaway | [email protected]
The first stage in the software pipeline is the retrieval of protein target sequences and protein template structures. Models are subsequently built by mapping each target sequence onto every available template structure using Modeller (http://salilab.org/modeller/). After filtering out non-unique models (based on an RMSD cutoff), the models are subjected to a two-stage molecular dynamics simulation refinement; the first with implicit solvation, the second with explicit solvation. For the explicit solvation step, the models are solvated such that each system contains the same number of waters. Finally, scripts are provided to package and compress the resulting models, ready for transfer or for set-up on other platforms such as Folding@Home.
scripts/ - Python wrapper scripts which accept parameters from the command-line or project metadata file
MSMSeeder/ - main code; can be used as a standalone Python library package
tests/ - to test whether the code is working correctly, run nosetests from this top-level directory
Clone this repository, move into this directory, then run:
python setup.py install
- OpenMM - https://simtk.org/home/openmm
- Modeller - http://salilab.org/modeller/
- mpi4py - http://mpi4py.scipy.org/
- mdtraj - http://mdtraj.org/
- PyMOL (optional, for model alignment/visualization) - http://www.pymol.org/
- Various other Python packages commonly used in scientific computing. Recommended aproach is to install either Enthought Canopy (https://www.enthought.com/products/canopy/) or Continuum Anaconda (https://store.continuum.io/)
The package can be used either by running the command-line scripts listed below, or by writing scripts to interact with the MSMSeeder Python library. The former approach is recommended for initially familiarizing yourself with the software. Each script can be run from the command-line with a '-h' flag, which will print information on their usage. They are intended to be run in the following order:
- InitMSMSeederProject.py
- GatherTargets.py
- GatherTemplates.py
- BuildModels.py
- RefineImplicitMD.py
- Solvate.py
- RefineExplicitMD.py
- PackageModels.py
The scripts have been parallelized with MPI where useful. Furthermore, the MD refinement steps use the OpenMM simulation package, which has particularly high performance on GPUs.
Some scripts will write certain parameters to a project metadata file (named "project-data.yaml" and stored in the project top-level directory). The purpose of this is partly to keep a record of the input parameters used, but if wanted, this file can be used to provide the same input parameters the next time the same script is run. Note that input parameters read from command-line arguments take precedence over those read from the project metadata file. A template project metadata file is provided in the source code.
Sometimes it is necessary to manually override certain automatic aspects of the target and template retrieval scripts. This can be achieved using a file named 'manual-specifications.yaml', to be placed in the project top-level directory (a template is provided in the source code). Currently, this allows one to manually specify individual domain spans for targets and templates retrieved from UniProt, as well as minimum and maximum acceptable domain lengths.
InitMSMSeederProject.py
GatherTargets.py --gather_from TargetExplorerDB --db_path ~/kinomeDB/database/database.xml
GatherTemplates.py --gather_from UniProt --uniprot_query 'domain:"Protein kinase" AND reviewed:yes' --uniprot_domain_regex '^Protein kinase(?!; truncated)(?!; inactive)' --structure_paths ~/pdbfiles ~/siftsfiles
BuildModels.py --targets ABL1_HUMAN_D0
RefineImplicitMD.py --targets ABL1_HUMAN_D0
Solvate.py --targets ABL1_HUMAN_D0
RefineExplicitMD.py --targets ABL1_HUMAN_D0
PackageModels.py --targets ABL1_HUMAN_D0 --package_for FAH
(run each script with '-h' flag for general info)
These two scripts are used to provide target data (ids and sequences) and template data (ids, sequences and structures). In each case, two different methods are provided, specified using the --gather_from flag. The 'TargetExplorerDB' method allows one to retrieve data from an existing TargetExplorer database (https://github.com/choderalab/targetexplorerdb), while the 'UniProt' method retrieves data from the UniProt web server. In the latter case, the --uniprot_query flag specifies the search string used to select UniProt entries. This uses the same syntax as the search function on the UniProt site; not all types of syntax may be supported, but most basic searches will work. While the initial data returned is in the form of UniProt entries (i.e. full-length proteins), the targets and templates selected by the scripts must be protein domains. Without further specification, the scripts will select all protein domains contained within the returned Uniprot entries. Often it will be necessary to select only a subset of these protein domains - e.g. if protein kinase domains are the desired target, then it will be necessary to filter out other domains contained in the full-length kinase proteins - and this can be achieved by supplying an appropriate regular expression along with the --uniprot_domain_regex flag.
It should also be noted that the output from these two scripts is simply a list of target sequences (targets/targets.fa), a list of template sequences (templates/templates.fa), and a set of template structures (templates/structures/[id].pdb). The subsequent model-building and refinement scripts can thus be run using any set of data specified in this way, allowing the GatherTargets.py and GatherTemplates.py scripts to be bypassed or modified if necessary. The main restrictions are that modified residues (e.g. phosphotyrosine) cannot be included, and that ids and residues in the templates.fa file must match with the template structure files.