PCC-FECalc: Binding Free Energy Calculations for Protein-Catalyzed Capture Agents

📖 Overview

This repository provides a python framework for performing binding free energy calculations for a user-defined target molecule and a protein-catalyzed capture (PCC) agent with a user-defined epitope sequence. The core of this package is based on "High-throughput virtual screening of protein-catalyzed capture agents for novel hydrogel-nanoparticle fentanyl sensors" and the corresponding repository. Please refer to the original paper for a more technical explanation of the theory and the setup behind the free energy calculations done with this framework.

For setup instructions, see Installation instructions.

🚀 Features

The framework is organized into four main submodules that work in serial to perform a complete binding free energy calculation using GROMACS:

  • PCCBuilder: This object creates a PCC with a given sequence, calculates GAFF2 parameters for it, and minimizes it.
  • TargetMOL: Similar to PCCBuilder, creates input structures and force field parameters for a provided target molecule.
  • FECalc: Brings together the PCC and the target molecule, solvates them in a water box, and performs minimization, NVT, and NPT equilibration, runs a parallel-bias metadynamics simulation, and reweights the resulting trajectory for free energy calculations.
  • postprocess: Uses the reweighted statistics to calculate the 3D binding free energy volume of the PCC-molecule complex and calculates the binding free energy and dissociation constant.

🛠 Usage

See example/pcc_submit_test.py for a sample calculation setup.

This framework additionally relies on PyMOL and acpype for PCC mutations and GAFF2 parameter generation. acpype is installed automatically as a Python package. PyMOL is required but not installed automatically because no pip wheel is available. Install PyMOL separately from pymol.org or build from the open-source GitHub repository. There's also a conda package package available:

conda install -c conda-forge pymol-open-source

Ensure the pymol executable is on your PATH.

The acpype workflow uses the sqm program from AmberTools, so install AmberTools and libgfortran5 after setting up the repository:

conda install -c conda-forge ambertools libgfortran5

The calculations happen through four steps and each step requires a JSON files with the necessary user parameters:

Step 1: Building the PCC

The settings file for this step is pre-made in FECalc/PCCBuilder_settings.JSON and generally requires no modification. By default, calling PCCBuilder.create will halt after each major stage so you can manually inspect the generated PCC structure and the parameters produced by acpype. Rerun the method to continue to the next step. If you prefer to run the entire preparation in one go, call PCCBuilder.create(check=False) to skip these pauses.

Step 2: Building the target

The settings file for this step should be created by the user. Two example are provided in example/ACT_settings.JSON for acetaminophen and example/FEN_settings.JSON for fentanyl. The mandatory entries are:

  • name: Name of the target. Used for creating subdirectories and making reports.
  • charge: Total charge of the target.
  • anchor1: Anchor point defined using the atoms on the target molecule. This is used together with anchor2 to define a vector that is used in determining the relative position and orientation of the target molecule with respect to the PCC during the PBMetaD calculations. See the original publication for a detailed explanation of how this vector is used in the collective variables.
  • anchor2: See above.
  • output_dir: Path to the folder to store the parameter calculations and minimization.
  • input_pdb_dir: Input pdb file of the target molecule structure.

Step 3: Enhanced sampling calculations

The settings file for this step should be created by the user. An example is provided in example/system_settings.JSON. The mandatory entries are:

  • PCC_output_dir: Path to the output folder that holds the PCC calculations.
  • PCC_settings_json: Path to the JSON file for the PCC.
  • MOL_settings_json: Path to the JSON file for the target.
  • temperature: Temperature of the simulations
  • box_size: Size of the simulation box. Cubic periodic.
  • complex_output_dir: Path to the out directory for the free energy calculations. The contents of this directory will be as follows:
{PCC sequence}_{target name}/
│-- em/               # Minimization
│-- nvt/              # NVT equilibration
│-- npt/              # NPT equilibration
│-- md/               # PBMetaD simulation
│-- reweight/         # Reweighting

The optional entries are:

  • scheduler: Scheduler used to allocate hardware resources. Supported options are "local" (default), "slurm", "pbs", and "lsf".
  • nodes, cores, threads: Hardware layout overrides for the number of nodes, cores per node, and threads per core. Defaults to 1 for each value; scheduler environment variables take precedence. These counts are used to set the correct -ntomp for GROMACS.
  • metad_settings: Parameters of the metadynamics simulation
  • n_steps: Number of steps for the metadynamics run. 2 fs step size. defaults to 800 ns.
  • metad_height: Height of the deposited Guassians. Defaults to 3.0 kJ/mol.
  • metad_pace: Pace of deposition. Defaults to 500 steps.
  • metad_bias_factor: Biasing factor for the PBMetaD bias. Defaults to 20.
  • postprocess_settings: Parameters for the post-processing and free energy calculations.
  • discard_initial: Initial duration of the PBMetaD simulation to discard for free energy calculations in ns. Defaults to 100 ns.
  • n_folds: Number of folds for block-analysis and uncertainty quantification. Defaults to 5.

Step 4: Free energy calculation

This step does not require a separate settings file as the parameters are set by the settings file in step 3. The post-processing step calculates the 3D integral of the bound basin and the unbound region and reports $\Delta G$ of binding in kJ/mol and $K_d$ in $\mu M$. The report is created in complex_output_dir/{PCC sequence}_{target name}/metadata.JSON.

📜 License

This project is licensed under the MIT License.