qFit is a collection of programs for modeling multi-conformer protein structures.
Electron density maps obtained from high-resolution X-ray diffraction data are a spatial and temporal average of all conformations within the crystal. qFit evaluates an extremely large number of combinations of sidechain conformers, backbone fragments and small-molecule ligands to locally explain the electron density.
If you use this software, please cite:
- Riley BT, Wankowicz SA, et al. qFit 3: Protein and ligand multiconformer modeling for X-ray crystallographic and single-particle cryo-EM density maps. Protein Sci. 30, 270–285 (2021)
- van Zundert, G. C. P. et al. qFit-ligand Reveals Widespread Conformational Heterogeneity of Drug-Like Molecules in X-Ray Electron Density Maps. J. Med. Chem. 61, 11183–11198 (2018)
- Keedy, D. A., Fraser, J. S. & van den Bedem, H. Exposing Hidden Alternative Backbone Conformations in X-ray Crystallography Using qFit. PLoS Comput. Biol. 11, e1004507 (2015)
We recommend using the conda package manager to install qFit.
You will need the following tools:
- git
- conda package manager (which you can get by installing Miniconda3)
Once these are installed, you can:
-
Create a new conda env & activate it
conda create --name qfit "python>=3.7,<3.9" conda activate qfit -
Install dependencies
conda install -c anaconda mkl numpy=1.20 conda install -c anaconda -c ibmdecisionoptimization \ cvxopt cplex -
Clone the latest release of the qFit source, and install to your conda env
git clone -b main https://github.com/ExcitedStates/qfit-3.0.git cd qfit-3.0 pip install .
-
You're now ready to run qFit programs! See usage examples below for some examples.
Unfortunately, the Anaconda repos don't contain 'osx-arm64' binaries for IBM's CPLEX and Intel's mkl.
We don't currently have plans to switch to a different MIQP solver (e.g. Gurobi).
As a workaround, you'll have to force conda to install the 'osx-64' binaries for everything (x86_64). macOS's Rosetta 2 translation will handle the Intel→AppleSilicon translation.
Instead of the first step in the above Installation section, use this:
- Create a new conda env & activate it
CONDA_SUBDIR=osx-64 conda create --name qfit "python>=3.8" conda activate qfit; conda env config vars set CONDA_SUBDIR=osx-64; conda deactivate conda activate qfit
then follow the rest of the instructions.
If you prefer to manage your environments using other methods, qFit has the following prerequisites:
Installation instructions using pip can be found in the docs folder.
Once dependencies are installed, you can clone the qFit source, and install to your env as above.
(Note: python setup.py install will only work if numpy has already been installed.)
qFit uses Black to format its code and provides a git hook to verify that code is properly formatted before allowing you to commit.
Before creating a commit, you will have to perform two actions:
- Install Black, either through a package manager or by running
python3 -m pip install --user black - Run
git config core.hooksPath .githooks/to use the provided pre-commit hook
The qfit package comes with several command line tools to model alternate
conformers into electron densities. You should select the command line tool that
is most suited for your task. Please, refer below for a few use case examples
to understand which tool is best suited for your needs.
To remove single-conformer model bias, qFit should be used with a composite omit map. One way of generating such map is using the Phenix software suite:
phenix.composite_omit_map input.mtz model.pdb omit-type=refine
An example test case (3K0N) can be found in the example directory. Additionally, you can find the Cryo-EM example (PDB: 7A4M) and the qFit-ligand example (PDB: 4L2L) in the example directory.
To model alternate conformers for all residues in a protein of interest using qFit, the following command should be used:
qfit_protein [COMPOSITE_OMIT_MAP_FILE] -l [LABELS] [PDB_FILE]
This command will produce a multiconformer model that spans the entirety of the input target protein. You will encounter two output files of interest in the directory where this command was run: multiconformer_model.pdb and multiconformer_model2.pdb.
If you wish to specify a different directory for the output, this can be done using the flag -d.
The multiconformer_model.pdb contains the output of running the qfit_residue routine for every protein residue. This file may contain up to five alternate conformers per residue. However, some of these conformers may be spurious as alternate conformers of neighboring residues are not consistent with regards to each other.
After calculating the conformers described in multiconformer_model.pdb, qfit_protein identifies consistent protein segments, where some of the spurious, overlapping conformers, are discarded. The final model, with the consistent multiconformer model is multiconformer_model2.pdb. This file should then be used as input to the refinement script provided in ./scripts.
By default, qFit expects the labels FWT,PHWT to be present in the input map. Different labels can be set accordingly using the flag -l.
Using the example 3K0N:
qfit_protein /path/to/3K0N.mtz -l 2FOFCWT,PH2FOFCWT /path/to/3K0N_refine.pdb
After multiconformer_model2.pdb has been generated, refine this model using:
qfit_final_refine_xray.sh /path/to/3K0N.mtz multiconformer_model2.pdb
Bear in mind that this final step currently depends on an existing installation of the Phenix software suite. This script is currently written to work with version Phenix 1.20.
qfit_residue [COMPOSITE_OMIT_MAP_FILE] -l [LABELS] [PDB_FILE] [CHAIN,RESIDUE]
Using the example 3K0N:
qfit_residue /path/to/3K0N.mtz -l 2FOFCWT,PH2FOFCWT /path/to/3K0N_refine.pdb A,113
This will produce a parsimonious model containing up to 5 alternate conformers for residue 113 of chain A of 3K0N.
qFit can also use ccp4 map files as input. To model alternate conformers using this type of map, it is also necessary to provide the resolution of the data, which can be achieved by using the flag -r.
qfit_protein [MAP_FILE] [PDB_FILE] -r [RESOLUTION]
For Cyro-EM ccp4 maps, you can use the example from the Apoferritin Chain A (PDB:7A4M)
qfit_protein /path/to/apoF_chainA.ccp4 /path/to/apoF_chainA.pdb -r 1.22
After multiconformer_model2.pdb has been generated, refine this model using:
qfit_final_refine_cryoem.sh /path/to/apoF_chainA.ccp4 apoF_chainA.pdb multiconformer_model2.pdb
Bear in mind that this final step currently depends on an existing installation of the Phenix software suite.
4. Deactivate backbone sampling and bond angle sampling to model alternate conformers for a single residue of interest (faster, less precise)
In its default mode, qfit_residue and qfit_protein samples backbone conformations using our KGS routine. This can be disabled using the --no-backbone flag.
For even faster (and less precise) results, one can also disable the sampling of the bond angle Cα-Cβ-Cγ, which can be deactivated by means of the --no-sample-angle flag.
Other useful sampling parameters that can be tweaked to make qFit run faster at the cost of precision:
- Increase step size (in degrees) of sampling around each rotamer: -s flag (default: 10)
- Decrease range/neighborhood of sampling about preferred rotamers: -rn flag (default: 60)
- Disable parsimonious selection of the number of conformers output by qFit using the Bayesian Information Criterion (BIC): --no-threshold-selection flag.
Using the example 3K0N:
qfit_residue /path/to/3K0N.mtz -l 2FOFCWT,PH2FOFCWT /path/to/3K0N_refine.pdb A,113 --no-backbone --no-sample-angle -s 20 -rn 45 --no-threshold-selection
For a full list of options, run:
qfit_residue -h
Using the example 3K0N:
qfit_protein /path/to/3K0N.mtz -l 2FOFCWT,PH2FOFCWT /path/to/3K0N_refine.pdb --no-backbone --no-sample-angle -s 20 -rn 45 --no-threshold-selection
The qfit_protein program can be executed in parallel and the number of concurrent processes can be adjusted using the -p flag.
Using the example 3K0N, spawning 30 parallel processes:
qfit_protein /path/to/3K0N.mtz -l 2FOFCWT,PH2FOFCWT /path/to/3K0N_refine.pdb -p 30
Depending on the resolution, the default parameters for the identification of consistent protein segments may prove too strict, leading to the removal of perfectly valid alternate conformers. We are currently working towards calibration of the parameters in a resolution-dependent manner.
In the meantime, if you notice that your conformer of interest is present in multiconformer_model.pdb, but was subsequently removed by qfit_protein, you can re-process the initial model with less stringent parameters using the qfit_segment program:
qfit_segment /path/to/3K0N.mtz -l 2FOFCWT,PH2FOFCWT /path/to/multiconformer_model.pdb --no-segment-threshold-selection -f 3
To model alternate conformers of ligands, the command line tool qfit_ligand should be used:
qfit_ligand [COMPOSITE_OMIT_MAP_FILE] -l [LABEL] [PDB_FILE] [CHAIN,LIGAND]
Using the example 4MS6:
qfit_ligand /path/to/ligand_composite_omit_map.mtz -l 2FOFCWT,PH2FOFCWT /path/to/4ms6.pdb A,702
We then recommend re-refining the output of qFit ligand along with the protein using:
qfit_final_refine_xray.sh 4l2l_qFit_ligand.pdb /path/to/4L2L.mtz
Where LIGAND corresponds to the numeric identifier of the ligand on the PDB (aka res. number).
The code is licensed under the MIT licence (see LICENSE).
Several modules were taken from the pymmlib package, originally licensed
under the Artistic License 2.0. See the licenses directory for a copy of the
original source code and its full license.
The elements.py is licensed under MIT, Copyright (c) 2005-2015, Christoph
Gohlke. See file header.
The Xpleo software and LoopTK package have been major inspirations for the inverse kinematics
functionality.