This Julia package implements the ExProSE algorithm that takes two protein structures and generates an ensemble of protein structures. The ensembles span conformational space and can be used to predict allosteric sites. The method is described in:

Greener JG, Filippis I and Sternberg MJE, Predicting protein dynamics and allostery using multi-protein atomic distance constraints, Structure, 2017, 25, 546-558. Link to paper.

The package also has some functions that may be of general use for structural bioinformatics including superimposition, RMSD, PCA, interaction finding in structures and distance geometry.

Install using add ProteinEnsembles from within the Julia package mode. Run using

exprose --i1 input_1.pdb --d1 input_1.dssp \
    --i2 input_2.pdb --d2 input_2.dssp \
    -n 50 -o exprose_out

where exprose is in the bin directory of the package.

Julia is required and can be downloaded here. Install ProteinEnsembles.jl by running add ProteinEnsembles from the Julia package REPL, which is entered by pressing ]. This will also automatically install a few other required Julia packages. If you want, the tests can be run using test ProteinEnsembles. If you wish to use the auto-parameterisation procedure (see below) you must also have TM-score installed.

To use ProteinEnsembles.jl you will need the following:

• PDB files of the protein of interest. Two is best, but one may be used (see the paper). They must have polar hydrogens only added; this can be done using tools such as Chimera or pdbtools. The chain labelling and residue numbering must be consistent between the files as this is used to find common atoms. Alternative atom locations are discarded. PDB files must also be a single model and not have any inserted residues. HETATM records are discarded by default.
• DSSP files corresponding to the PDB files above. These can be obtained using dssp.

These instructions are tailored towards Mac/Unix. However they could be modified to work on Windows.

Although organised as a Julia package, ProteinEnsembles.jl is primarily designed for use from the command line. The exprose script in the bin directory implements this. For example, to see the command line options run

~/.julia/packages/ProteinEnsembles/xxxxx/bin/exprose -h

where xxxxx is the directory of the latest package version. For easy access to the exprose command you might like to add the following line to your profile:

export PATH=$PATH:~/.julia/packages/ProteinEnsembles/xxxxx/bin

Then, if all input files are in your current directory, run the program as follows:

# Generate an ensemble of 50 structures with an output directory exprose_out
exprose --i1 input_1.pdb --d1 input_1.dssp --i2 input_2.pdb \
    --d2 input_2.dssp -n 50 -o exprose_out

# Use a tolerance weighting of 0.5
exprose --i1 input_1.pdb --d1 input_1.dssp --i2 input_2.pdb \
    --d2 input_2.dssp -n 50 -o exprose_out -w 0.5

# Generate an ensemble from a single structure with a tolerance weighting of 1.0
exprose --i1 input_1.pdb --d1 input_1.dssp -n 50 -o exprose_out -w 1.0

The method may also be run from within Julia. The below Julia script does the same thing as the first example above:

using ProteinEnsembles
runpipeline(
    i1="input_1.pdb",
    d1="input_1.dssp",
    i2="input_2.pdb",
    d2="input_2.dssp",
    n_strucs=50,
    out_dir="exprose_out"
)

Or, to split it up a little into the constituent functions:

using ProteinEnsembles
constraints_com, constraints_one, constraints_two = interactions(
    "input_1.pdb",
    "input_1.dssp",
    "input_2.pdb",
    "input_2.dssp"
)
ensemble_com = generateensemble(constraints_com, 50)
makedirectories("exprose_out", ["pdbs", "pymol", "pcs"])
runanalysis("exprose_out", ensemble_com, constraints_one, constraints_two)

The auto-parameterisation procedure can select a more suitable tolerance weighting value (see the paper). TM-score must be installed to do this. For example:

exprose-param --i1 input_1.pdb --d1 input_1.dssp --i2 input_2.pdb \
    --d2 input_2.dssp -o exprose_param -t TMscore

runs the auto-parameterisation procedure with the -t option specifying the command to run TM-score. The last line of the output gives a suggested tolerance weighting. This value is also written out to suggested.tsv. Use this value in a normal exprose run as above.

To predict allosteric sites you should run LIGSITEcs on the second input structure (the one you give as --i2). You then need to run the cluster-ligsite script in bin to assign the points to pockets:

cluster-ligsite pocket_r.pdb pocket_all.pdb pocket_points.pdb

where pocket_r.pdb and pocket_all.pdb are in the LIGSITEcs output. Then carry out an exprose run with the pocket_points.pdb file (-l) and the number of pockets (e.g. top 4) to perturb at (-m) as parameters:

exprose --i1 input_1.pdb --d1 input_1.dssp --i2 input_2.pdb \
    --d2 input_2.dssp -n 50 -o exprose_out -l pocket_points.pdb -m 4

A tolerance weighting from an auto-parameterisation run can also be used here. View the predictions.tsv output file to get the order of allosteric pocket predictions. Note that other pocket prediction software can be used provided you can get the output into the same format as pocket_points.pdb, i.e. pocket cavity points with the pocket number in the residue number column.

The output directory contains the following:

• input_1.pdb and input_2.pdb: atoms used from the input structures are written back out and superimposed.
• pdbs: generated structures in PDB format. Superimposed to input_1.pdb and input_2.pdb.
• pcs: projections onto the principal components (PCs) from the principal component analysis of the generated structures. Contains files for generated (pcs.tsv) and input structures (pcs_input_1.tsv and pcs_input_2.tsv) - line n corresponds to structure n and column c corresponds to PC c. Has graphs of these for the first few PCs (pc_x_y.png). Also includes a list of PCs ordered by decreasing distance between the input structures (pcs_input_dist.tsv) and the percentage variation explained by each PC (evals_spread.tsv).
• pymol: PyMol scripts to view PCs on input_1.pdb, e.g. run pymol input_1.pdb pymol/view_pc_1.pml.
• rmsds_input_1.tsv and rmsds_input_2.tsv: RMSDs of generated structures to the input structures. Line n corresponds to structure n.
• rmsfs.tsv and rmsfs.png: RMSFs of each residue over the ensemble of generated structures, and a plot of this. Line n corresponds to residue index n.
• spe_scores.tsv: SPE error scores of generated structures (see paper). Line n corresponds to structure n.

For allosteric site prediction there will be pdbs_mod_n and mod_n containing similar information for each perturbed ensemble, as well as the ratio of RMSF values to the unperturbed ensemble (rmsfs_ratio.tsv). There will also be the order of allosteric predictions (predictions.tsv) and the size of the perturbation on modulating each site (perturbations.tsv), which is the RMSD between the centroid structure of the perturbed and unperturbed ensembles.

The default plot colours are blue for generated structures, red for input structure 1, green for input structure 2 and orange for perturbed ensemble structures.

The results from the paper can be generated using the instructions in paper_results. Information on the ensemble and allosteric datasets is in datasets.

ExProSE can generate 250 structures in ~20 minutes for T4-lysozyme (162 residues) on a 3.1 GHz Intel Core i7 processor.

If you find any bugs in the software or have a comment or feature request, please open an issue on GitHub or email Joe Greener (j.greener at ucl.ac.uk).

• At some point the download link to LIGSITEcs seems to have died. You may need to contact the authors to get the software.
• All default values for parameters used in the code can be found and modified in src/defaults.jl.
• Auto-parameterisation works fine on all OSs but the auto-parameterisation tests are disabled by default to make the CI build pass. If you want to run the parameterisation tests, set run_param_test in test/runtests.jl to true.
• Julia utilities to deal with protein structures and PDB files can be found in BioStructures.jl and MIToS.jl.
• ExProSE users might also like to try tCONCOORD and NMSim.