Scope

This repository contains source code for the classical docking setup, input data processing, result generation, and evaluation on PLINDER and PoseBusters protein-ligand docking benchmarks.

Details: ArtiDock: Accurate Machine Learning Approach to Protein–Ligand Docking Optimized for High-Throughput Virtual Screening

Dataset

The ready-to-use datasets can be downloaded from Zenodo:

• PLINDER (plinder-v2-eval.tar.gz)
• PoseBusters (posebusters_paper_data.zip)

Setup

1. Install Miniconda or Anaconda.

2. Clone the source code and set up the environment.

   git clone https://github.com/receptor-ai/dock-eval.git
   cd dock-eval/
   

3. Create environment.

   • AutoDock:

   conda env create --file requirements/autodock/environment.yml
   conda activate autodock
   pip install -r requirements/autodock/requirements.txt --no-cache-dir
   

   If you are going to use AutoDock-GPU, it should be installed additionally. Run conda activate autodock && bash requirements/autodock/install_autodock_gpu.sh. The script was tested on a machine with an RTX 3000/4000 Series GPUs and CUDA 12.4 using nvidia/cuda:12.4.1-devel-ubuntu22.04 Docker image template. You may also try conda install hcc::autodock-gpu.

   • Glide:

   To use Glide, you should have Schrödinger Suite installed on your machine. Locate the path that contains utilities/prepwizard, ligprep, and glide tools required for the docking (e.g., /home/user/software/schrodinger2024-3/).

   • Evaluation:

   conda env create --file requirements/eval/environment.yml
   conda activate dock-eval
   pip install -r requirements/eval/requirements.txt --no-cache-dir
   

Usage

• The short custom usage examples of the docking module are available in examples/docking.ipynb (autodock environment)
• The short custom usage example of the evaluation module is available in examples/evaluate.ipynb (dock-eval environment)

Running docking for benchmark

• AutoDock:

  conda activate autodock
  
  # AutoDock-CPU (AutoDock4) example:
  python -m autodock.plinder_v2 --ds_root ../data/plinder-v2-eval/ --split test --workdir workdir/autodock-cpu/plinder-v2-test --docked_sdf result_1.sdf --grid_size 20 20 20 --type cpu --params ga_num_evals=350000 ga_run=20 --parallel
  
  # AutoDock-GPU example:
  python -m autodock.plinder_v2 --ds_root ../data/plinder-v2-eval/ --split test --workdir workdir/autodock-gpu/plinder-v2-test --docked_sdf result_1.sdf --grid_size 20 20 20 --type gpu
  
  # Vina example:
  python -m autodock.plinder_v2 --ds_root ../data/plinder-v2-eval/ --split test --workdir workdir/autodock-vina/plinder-v2-test --docked_sdf result_1.sdf --grid_size 20 20 20 --type vina --params exhaustiveness=16
  

• Glide:

  export SCHRODINGER="/home/user/software/schrodinger2024-3/"
  
  # only basic dependencies are needed, so we can use the already installed dock-eval
  conda activate dock-eval
  
  python -m glide.plinder_v2 --ds_root ../data/plinder-v2-eval/ --split test --workdir workdir/glide/plinder-v2-test --docked_sdf result_1.sdf --outerbox 20 20 20 --dock_params POSTDOCK_NPOSE=80 --parallel
  

These examples will launch docking on the PLINDER test dataset (holo) and save results to --workdir. To launch docking on the PLINDER MLSB test dataset (undound), change --split to test-mlsb and provide another --workdir. To launch docking on the PoseBusters dataset, use the corresponding posebusters.py scripts, remove --split argument, and provide the --ds_root folder with complexes (e.g., python -m autodock.posebusters --ds_root ../data/posebusters_benchmark_set/ --workdir workdir/autodock-cpu/posebusters/ --docked_sdf result_1.sdf --grid_size 25 25 25 --type cpu --parallel).

Running evaluation for benchmark

To evaluate the docking performance, use the same --ds_root, --split (for PLINDER), --workdir, and --docked_sdf arguments as for a docking launch. Example for AutoDock-CPU evaluation:

conda activate dock-eval

python -m eval.plinder_v2 --ds_root ../data/plinder-v2-eval/ --split test --workdir workdir/autodock-cpu/plinder-v2-test --docked_sdf result_1.sdf --parallel

TODO

• Virtual screening mode for docking
• Unit tests