Relative Free Energies with the OpenFE CLI#

This tutorial will show how to use the OpenFE command line interface to get free energies – with no Python at all! This will work for simple setups, you may need to use the Python interface for more complicated setups.

The entire process of running the campaign of simulations is split into 3 stages; each of which corresponds to a CLI command:

  1. Setting up the necessary files to describe each of the individual simulations to run.

  2. Running the simulations.

  3. Gathering the results of separate simulations into a single table.

To work through this tutorial, start out with a fresh directory. You can download the tutorial materials (including this file) using the command:

openfe fetch rbfe-tutorial

Then when you run ls, you should see that your directory has this file, cli_tutorial.md, a notebook called python_tutorial.ipynb, and files with the molecules we’ll use in this tutorial: tyk2_ligands.sdf and tyk2_protein.pdb.

Setting up the campaign#

The CLI makes setting up the simulation very easy – it’s just a single CLI command. There are separate commands for relative binding free energy (RBFE) and relative hydration free energy setups (RHFE).

For RBFE campaigns, the relevant command is openfe plan-rbfe-network. For RHFE, the command is openfe plan-rhfe-network. They work mostly the same, except that the RHFE planner does not take a protein. In this tutorial, we’ll do an RBFE calculation. The only difference for RBFE is in the setup stage – running the simulations and gathering the results are the same.

To run the command, we do the following:

  • Read all the ligands from the SDF by giving the option -M tyk2_ligands.sdf. You can also use -M with a directory, and it will load all molecules found in any SDF or MOL2 file in that directory.

  • Pass a PDB of the protein target (TYK2) with -p tyk2_protein.pdb.

  • Instruct openfe to ouput files into a directory called network_setup with the -o network_setup option.

openfe plan-rbfe-network -M tyk2_ligands.sdf -p tyk2_protein.pdb -o network_setup

Planning the campaign may take some time, as it tries to find the best network from all possible transformations. This will create a directory called network_setup, which is structured like this:

network_setup
├── ligand_network.graphml
├── network_setup.json
└── transformations
    ├── easy_rbfe_lig_ejm_31_complex_lig_ejm_42_complex.json
    ├── easy_rbfe_lig_ejm_31_complex_lig_ejm_46_complex.json
    ├── easy_rbfe_lig_ejm_31_complex_lig_ejm_47_complex.json
    ├── easy_rbfe_lig_ejm_31_complex_lig_ejm_48_complex.json
    ├── easy_rbfe_lig_ejm_31_complex_lig_ejm_50_complex.json
    ├── easy_rbfe_lig_ejm_31_solvent_lig_ejm_42_solvent.json
    ├── easy_rbfe_lig_ejm_31_solvent_lig_ejm_46_solvent.json
[continues]

The ligand_network.graphml file describes the atom mappings between the ligands. We can visualize it with the openfe view-ligand-network command:

openfe view-ligand-network network_setup/ligand_network.graphml

This opens an interactive viewer. You can move the ligand names around to get a better view of the structure, and if you click on the edge, you will see the mapping for that edge.

The files that describe each individual simulation we will run are located in the transformations subdirectory. Each JSON file represents a single alchemical leg to run, and contains all the necessary information to run that leg. A single RBFE between a pair of ligands requires running two legs of an alchemical cycle (JSON files): one for the ligand in solvent, and one for the ligand complexed with the protein. The results from these two simulations can then be combined to obtained a single \(\Delta\Delta G\) relative binding free energy value. Filenames indicate ligand names as taken from the SDF; for example, the file easy_rbfe_lig_ejm_31_complex_lig_ejm_42_complex.json is the leg associated with the tranformation of the ligand lig_ejm_31 into lig_ejm_42 while in complex with the protein.

Note that this specific setup makes a number of choices for you. All of these choices can be customized in the Python API. Here are the specifics on how these simulation are set up:

  1. LOMAP is used to generate the atom mappings between ligands, with a 20-second timeout, core-core element changes disallowed, and max3d set to 1.

  2. The network is a minimal spanning tree, with the default LOMAP score used to score the mappings.

  3. Solvent is water with NaCl at an ionic strength of 0.15 M (neutralized) with a minimum distance of 1.2 nm from the solute to the edge of the box.

  4. The protocol used is OpenFE’s OpenMM-based Hybrid Topology RFE protocol, with default settings.

Running the simulations#

For this tutorial, we have precalculated data that you can load, since running the simulations can take a long time. However, you could, in principle, run each simulation on your local machine.

You can run each leg individually by using the openfe quickrun command. It takes a transformation JSON as input, and the flags -o to give the final output JSON file and -d for the directory where simulation results should be stored. For example,

openfe quickrun path/to/transformation.json -o results.json -d working-directory

where path/to/transformation.json is the path to one of the files created above.

When running a complete network of simulations, it is important to ensure that the file name for the result JSON and name of the working directory are different for each leg, otherwise you’ll overwrite results. We recommend doing that with something like the following, which uses the fact that the JSON files in network_setup/transformations/ have unique names, and creates directories and result JSON files based on those names. To run all legs sequentially (not recommended) you could do something like:

# this will take a very long time! don't actually do it!
for file in network_setup/transformations/*.json; do
  relpath=${file:30}  # strip off "network_setup/transformations/"
  dirpath=${relpath%.*}  # strip off final ".json"
  openfe quickrun $file -o results/$relpath -d results/$dirpath
done

In practice, you probably want to submit these to a queue. In that case, you’ll want to create a new job script for each simulation JSON file, and the core of that job script will be to run the openfe quickrun command above.

Details of what information is needed in that job script will depend on your computing center. Here is an example of a very simple script that will create and submit a job script for the simplest SLURM use case:

for file in network_setup/transformations/*.json; do
  relpath=${file:30}  # strip off "network_setup/transformations/"
  dirpath=${relpath%.*}  # strip off final ".json"
  jobpath="network_setup/transformations/${dirpath}.job"
  cmd="openfe quickrun $file -o results/$relpath -d results/$dirpath"
  echo -e "#!/usr/bin/env bash\n${cmd}" > $jobpath
  sbatch $jobpath
done

Note that the exact structure of the results directory is not important, as long as all result JSON files are contained within a single directory tree. The approach listed here is what was used for the example results that we’ll download in the next section.

Gathering the results#

To get example data, use the following commands:

openfe fetch rbfe-tutorial-results
tar xzf rbfe_results.tar.gz

This will create a directory called results/ that contains files with the file structure you would get from running the calculations as above. The result JSON files are the actual results of a simulation. Other files that are generated during the simulation (such as detailed simulation information) have been replaced by empty files to keep the size smaller. The structure looks something like this:

results
├── easy_rbfe_lig_ejm_31_complex_lig_ejm_42_complex
│   ├── shared_RelativeHybridTopologyProtocolUnit-3ea82011-75f0-4bb6-b415-e7d05bd012f6
│   │   ├── checkpoint.nc
│   │   └── simulation.nc
│   ├── shared_RelativeHybridTopologyProtocolUnit-5262feb6-cb50-4bb2-90a2-359810c2bb9c
│   │   ├── checkpoint.nc
│   │   └── simulation.nc
│   └── shared_RelativeHybridTopologyProtocolUnit-7a6def34-2967-4452-8d47-483bc7219c06
│       ├── checkpoint.nc
│       └── simulation.nc
├── easy_rbfe_lig_ejm_31_complex_lig_ejm_42_complex.json
├── easy_rbfe_lig_ejm_31_complex_lig_ejm_46_complex
│   ├── shared_RelativeHybridTopologyProtocolUnit-ad113e55-5636-474e-9be3-ee77fe887e77
│   │   ├── checkpoint.nc
│   │   └── simulation.nc
│   ├── shared_RelativeHybridTopologyProtocolUnit-ca74ad3c-2ac8-4961-be7c-fa802a1ec76b
│   │   ├── checkpoint.nc
│   │   └── simulation.nc
│   └── shared_RelativeHybridTopologyProtocolUnit-f848e671-fdd3-4b8d-8bd2-6eb5140e3ed3
│       ├── checkpoint.nc
│       └── simulation.nc
├── easy_rbfe_lig_ejm_31_complex_lig_ejm_46_complex.json
[continues]

The JSON results file contains not only the calculated \(\Delta G\), and uncertainty estimate, but also important metadata about what happened during the simulation. In particular, it will contain information about any errors or failures that occurred – these errors will not cause the entire campaign to fail, and will be recorded so you can later analyze what went wrong.

To gather all the \(\Delta G\) estimates into a single file, use the openfe gather command from within the working directory used above:

openfe gather ./results/ -o final_results.tsv

This will write out a tab-separated table of results, including both the \(\Delta G\) for each leg and the \(\Delta\Delta G\) computed from pairs of legs. The first column is a description of the data, e.g., DGcomplex(ligandB, ligandA) for the \(\Delta G\) of the transformation of ligand A into ligand B complexed to a protein, or DDGbind(ligandB, ligandA) for the binding \(\Delta\Delta G\) going from ligand A to ligand B. The second column tells the type of the result, either RBFE for a relative result or solvent/complex for an individual leg. The next two columns are the labels of the ligands, and then the computed result and its uncertainty.

The resulting file looks something like this:

measurement  type    ligand_i    ligand_j    estimate (kcal/mol) uncertainty (kcal/mol)
DDGbind(lig_ejm_48, lig_ejm_31) RBFE    lig_ejm_31  lig_ejm_48  0.45    0.17
DDGbind(lig_jmc_28, lig_ejm_46) RBFE    lig_ejm_46  lig_jmc_28  -0.12   0.044
DDGbind(lig_ejm_46, lig_ejm_31) RBFE    lig_ejm_31  lig_ejm_46  -0.73   0.097
DDGbind(lig_ejm_50, lig_ejm_31) RBFE    lig_ejm_31  lig_ejm_50  0.94    0.072
DDGbind(lig_ejm_42, lig_ejm_31) RBFE    lig_ejm_31  lig_ejm_42  0.49    0.09
DDGbind(lig_jmc_23, lig_ejm_46) RBFE    lig_ejm_46  lig_jmc_23  -0.39   0.046
DDGbind(lig_ejm_43, lig_ejm_42) RBFE    lig_ejm_42  lig_ejm_43  1.2 0.14
DDGbind(lig_jmc_27, lig_ejm_46) RBFE    lig_ejm_46  lig_jmc_27  -0.65   0.1
DDGbind(lig_ejm_47, lig_ejm_31) RBFE    lig_ejm_31  lig_ejm_47  0.016   0.15
DGsolvent(lig_ejm_31, lig_ejm_48)   solvent lig_ejm_31  lig_ejm_48  -20.0   0.043
DGcomplex(lig_ejm_31, lig_ejm_48)   complex lig_ejm_31  lig_ejm_48  -19.0   0.17
DGsolvent(lig_ejm_46, lig_jmc_28)   solvent lig_ejm_46  lig_jmc_28  14.0    0.043
DGcomplex(lig_ejm_46, lig_jmc_28)   complex lig_ejm_46  lig_jmc_28  14.0    0.0069
[continues]