Beta version of GalaxyDock-DL for specialized use (Original version: [email protected]:seoklab/galaxydock_dl.git)

GalaxyDock-DL is a protein-ligand docking method which utilizes Conformational Space Annealing(CSA) as a sampling algorithm and deep learning-based scoring functions.

Installation Guide (linux only)

1. Clone this Git repository
   

$ git clone [email protected]:seoklab/colab_gd_dl.git

Below commands should be run in terminal from main directory location.

2. Install torch geometric files. (you can ignore errors)
   

$ pip install torch_geometric
$ pip install torch_scatter

3. Install source files (gd_dl)
   

$ pip install -e .

Usage (linux only)

Below commands should be run in terminal from main directory location.
We recommend checking src/gd_dl/path_setting.py if you want to change path settings.

Running docking

Run docking for a single ligand mol2 file and a protein receptor file without the ligand. (A center coordinate of a docking box (22.5 angstrom^3) is usually set to a coordinate of cognate ligand's geometric center for docking box to include a binding site.)

Random seeds were set to zero, but you can modify random seeds by adding argument

--random_seed <random_seed value>

Default output directory is set to current working directory, but you can modify random seeds by adding argument

--out_dir <location of output directory>

You can change length of docking box by adding argument

--box_size <box size value in angstrom>

$ python scripts/run_gd_dl.py -p <path to protein receptor file(.pdb)> -l <path to ligand file(.mol2)> -x <center x coordinate of a docking box> -y <center y coordinate of a docking box> -z <center z coordinate of a docking box>

Example for 3rsx

$ python scripts/run_gd_dl.py -p ./example/3rsx/3rsx_protein.pdb -l ./example/output_dir/charged_ligand.mol2 -x 69.637 -y 49.989 -z 10.160 --out_dir example/output_dir/

If you want to run docking in terminal from a different directory, you can use bash command with '-d ' below

$ python scripts/run_gd_dl_from_other_directory -d <Path to main directory> -p <path to protein receptor file(.pdb)> -l <path to ligand file(.mol2)> -x <center x coordinate of a docking box> -y <center y coordinate of a docking box> -z <center z coordinate of a docking box>

Output files

• GalaxyDock_fb.mol2: Contains the final output ligand poses, sorted by total score.
• GalaxyDock_fb.E.info: Provides the scores of the final output ligand poses in the final bank, sorted by total score.

For GalaxyDock_fb.E.info:

• The second column (Energy) shows the ranking scores of output poses inferred by neural network scoring functions.
• You can ignore the values in the l_RMSD column, as they only represent RMSD calculated by the Hungarian algorithm between processed input ligand poses and output ligand poses.
• You can also ignore the other columns, which correspond to the values of GalaxyDock BP2 Score energy components multiplied by their weights (ATDK_E: AutoDock Energy, INT_E: AutoDock intra-ligand energy, DS_E: Drug Score, HM_E: Hydrophobic interaction, PLP: PLP score).

GalaxyDock_ib.mol2: Initial ligand conformations in the first bank
box.pdb: Representation of docking box
GalaxyDock_cl.mol2: clustered final output ligand poses sorted by total score

Other output files are used during initialization or sampling and not important after docking is finished.

You can view ligand conformations directly using UCSF chimera
For example,

$ chimera GalaxyDock_fb.mol2

or you can view ligand conformations and protein receptor

$ chimera GalaxyDock_fb.mol2 <path to protein receptor file(.pdb)>

Citation

If you utilize this code or the models in your research, please cite the following paper:

@article{lee2024galaxydock,
  title={GalaxyDock-DL: Protein--Ligand Docking by Global Optimization and Neural Network Energy},
  author={Lee, Changsoo and Won, Jonghun and Ryu, Seongok and Yang, Jinsol and Jung, Nuri and Park, Hahnbeom and Seok, Chaok},
  journal={Journal of Chemical Theory and Computation},
  year={2024},
  publisher={ACS Publications}
}

License

All code is licensed under the MIT license. The weights of the neural networks are licensed under the CC BY-NC 4.0 license.