ProDock

Automatic pipeline for molecular modeling

Overview

ProDock is a toolkit for building automated molecular docking workflows. It is designed for campaigns involving multiple receptors, multiple ligands, and multiple docking engines, with support for downstream pose extraction, interaction profiling, visualization, and SQLite-backed result management. Please visit Documentation for more details.

Documentation map

The project aims to provide one consistent workflow for:

• receptor and ligand preparation
• batch docking across one or more engines
• pose extraction into standardized tables
• interaction analysis from docked complexes
• database storage for poses, scores, and interactions
• reproducible downstream analysis

This makes ProDock useful both for small docking experiments and for larger benchmark-style or screening-style studies.

In addition to classical scoring-function engines (e.g. AutoDock Vina), ProDock provides a semi-automated, rank-resolved, multi-engine pipeline that combines global docking with DiffDock and local docking with GNINA for virtual screening. See DiffDock + GNINA GPU pipeline below.

Main capabilities

Docking workflow

ProDock supports automated docking workflows across multiple receptor-ligand combinations and can be organized as single-target, multi-ligand, or multi-receptor campaigns.

Typical use cases include:

• one receptor with many ligands
• many receptors with one ligand set
• many receptors with many ligands
• comparison of multiple docking engines on the same campaign

Post-processing

After docking, ProDock can parse docking outputs and convert them into structured pose tables with canonical columns such as:

• receptor_id
• ligand_id
• engine
• pose_rank
• affinity
• mol
• pose_id

These standardized tables make it easier to compare poses across engines and campaigns.

Interaction analysis

ProDock supports protein-ligand interaction extraction and summarization, enabling residue-level interaction profiles for each pose. This is intended for downstream comparison, ranking, and interpretation of docking results.

Rank-resolved multi-engine scoring

The DiffDock + GNINA pipeline preserves multiple poses per ligand instead of reducing each docking run to a single top-ranked conformer. Each pose rank is evaluated independently using engine-specific scores and additional descriptors.

For GNINA, these descriptors include:

• empirical affinity;
• CNNpose;
• CNNaffinity;
• Reference-based protein–ligand interaction-fingerprint similarity: Similarity-type1 and Similarity-type2;
• Steric clash counts; and
• Optional electrostatic solvation energy.

For DiffDock, the evaluated descriptors include:

• Confidence score;
• Unoccupied region, representing the absence of an interaction surface between the protein and ligand; and
• Percentage of ligand atoms remaining within the binding site and percentage of Occupancy.

Database-backed storage

ProDock includes SQLite-based storage to keep docking poses and interaction records in a structured and queryable form. This is especially useful when handling many receptors, ligands, docking engines, and poses.

Database architecture

ProDock stores docking outputs and associated interaction information in a relational SQLite database. This supports scalable querying, reproducible analysis, and easy export into pandas dataframes.

Database architecture figure:

This architecture is intended to support:

• pose-centric storage
• stable pose identifiers
• interaction lookup by pose, receptor, ligand, or engine
• multi-receptor and multi-engine benchmarking workflows
• clean integration with pandas- and RDKit-based analysis

Installation

ProDock requires Python 3.11 or later. For a source installation:

git clone --recurse-submodules \
  https://github.com/Medicine-Artificial-Intelligence/ProDock.git
cd ProDock

conda create --name prodock-env python=3.11
conda activate prodock-env
python -m pip install -e .

Choose an additional profile only when needed:

Verify the package and inspect the command-line interface:

python -c "import prodock; print(prodock.__version__)"
prodock --help

Docking engines are external executables. Install the engines named in your campaign and ensure commands such as vina, smina, qvina, or qvina-w are available on PATH.

DiffDock + GNINA GPU pipeline

The rank-resolved multi-engine pipeline builds on established cheminformatics, visualization, and molecular-simulation libraries, including RDKit, PyMOL, Open Babel, OpenMM, MDAnalysis, ProLIF, Biopython, and APBS. It writes structured intermediate files and target-specific output directories so that protein preparation, ligand preparation, docking, re-ranking, and downstream inspection can be repeated from explicit inputs.

Requirements

The GPU pipeline requires:

• Linux
• Git
• Conda
• Python 3.11
• NVIDIA GPU recommended
• A compatible CUDA installation for GPU-enabled DiffDock and GNINA execution

1. Clone ProDock with submodules

Clone the repository together with its registered submodules:

git clone \
  --recurse-submodules \
  https://github.com/Medicine-Artificial-Intelligence/ProDock.git

cd ProDock

The submodule directories should be located under Project/:

ProDock/
├── Project/
│   ├── DiffDock/
│   └── gnina/
├── environment.yml
└── README.md

2. Create the ProDock environment

Create the Conda environment from environment.yml:

conda env create --file environment.yml

Activate the environment:

conda activate ProDock

3. Install DiffDock

Follow the DiffDock installation instructions. Make sure to clone DiffDock into Project/.

git clone \
  https://github.com/gcorso/DiffDock.git \
  Project/DiffDock

4. Install ESM

The ESM Python package must be located directly inside the DiffDock directory:

ProDock/Project/DiffDock/esm/

The final directory structure must be:

Project/DiffDock/
└── esm/
    ├── esmfold/
    ├── inverse_folding/
    ├── model/
    ├── __init__.py
    ├── data.py
    ├── modules.py
    ├── pretrained.py
    └── ...

Clone the ESM repository into a temporary directory:

git clone \
  https://github.com/facebookresearch/esm.git \
  esm_source

Copy the inner ESM Python package directly into Project/DiffDock/esm:

cp -r esm_source/esm Project/DiffDock/esm

5. Install GNINA

The GNINA directory must be located under Project/:

ProDock/Project/gnina/

The GNINA executable must be located at:

ProDock/Project/gnina/gnina

Enter the GNINA directory:

mkdir -p Project/gnina
cd Project/gnina

Download a compatible GNINA Linux binary from the official release page.

Using wget:

wget -O gnina "<GNINA_BINARY_URL>"

Alternatively, using curl:

curl -L "<GNINA_BINARY_URL>" -o gnina

Replace <GNINA_BINARY_URL> with the URL of the selected GNINA release asset.

Make the binary executable:

chmod +x gnina

Return to the ProDock root directory:

cd ../..

Test the GNINA installation:

./Project/gnina/gnina --version

Display the available GNINA options:

./Project/gnina/gnina --help

6. Complete installation

The final directory structure should resemble:

ProDock/
├── Project/
│   ├── Analysis_script/
│   ├── Docking_script/
│   ├── Optimization_script/
│   ├── Preparation_script/
│   ├── DiffDock/
│   │   ├── esm/
│   │   └── ...
│   └── gnina/
│       ├── gnina
│       └── ...
├── Data/
│   └── fig/
├── environment.yml
├── LICENSE
└── README.md

Overall workflow

Protein Preparation

Ligand Preparation

Reproducing the Case Study

To reproduce the case study, first clone and install ProDock as described above. Then run the command-line workflow using the provided configuration files:

python -m prodock \
  --config Data/case/config.json \
  --receptor-json Data/case/receptor.json \
  --ligand-json Data/case/ligand.json

This command executes the case-study docking workflow using:

• Data/case/config.json: global workflow and docking settings
• Data/case/receptor.json: receptor definitions
• Data/case/ligand.json: ligand definitions

Pose Re-ranking Optimization and Analysis

The scripts under Project/Optimization_script/ merge per-pose GNINA and DiffDock results and use Optuna to tune descriptor thresholds for ROC-AUC, PR-AUC, or logAUC. Install their optional dependencies from the repository root:

pip install -e ".[reranking]"

The repository already contains the 43 GNINA/DiffDock target pairs used by these scripts:

Project/benchmark/
├── result_gnina/{target}_final.csv
└── result_diffdock/{target}_final.csv

These are DUD-E-style files without an embedded activity column. Pass --dude-labels to apply the explicit dataset convention that ZINC* identifiers are decoys and all other identifiers are actives. The MT1 pair contains NONE as its only non-ZINC identifier and should be excluded from aggregate results until a valid active record is recovered.

Optimize one target:

python Project/Optimization_script/optuna_combine_all_structure.py \
  --protein ABL1 \
  --base-dir Project/benchmark \
  --dude-labels \
  --metric roc-auc

Useful options include:

• --scoring-metric {affinity,cnnaffinity,cnn-combined}
• --metric {roc-auc,pr-auc,logauc}
• --n-trials, --n-jobs, and --top-k
• --split {train,test}, --data-dir split_merged, and --eval-on-test
• --activity-column for datasets that carry explicit binary labels

Run all targets discovered under --base-dir:

python Project/Optimization_script/run_all_proteins_optimization.py \
  --base-dir Project/benchmark \
  --dude-labels \
  --exclude-proteins MT1 \
  --metric roc-auc \
  --parallel 4

The original nested export layout remains supported:

all/
├── gnina/{target}/Confidence_score/{target}_{split}_final.csv
└── diffdock/{target}/Confidence_score/{target}_{split}_final.csv

For the training split, the legacy unsplit filename {target}_final.csv is also accepted.

Results default to results_all_structure/{target}/ relative to the working directory. Use --output-dir to choose another location. Optimization output, Optuna studies, extracted threshold tables, and publication plots are generated artifacts and are intentionally ignored by Git.

Extract thresholds from the nine scoring/metric configuration directories:

python Project/Analysis_script/extract_optimized_thresholds.py \
  --root /path/to/configuration-results \
  --output-dir threshold_csv

To generate the publication plots, place the nine summary JSON files (aff_roc.json, cnn_pr.json, combined_log.json, and the other scoring/metric combinations) beside visualize_results.py, then run:

python Project/Analysis_script/visualize_results.py

See REPRODUCIBILITY.md for the tracked-input manifest, artifact policy, and end-to-end verification commands.

Setting Up Your Development Environment

Create the development environment and start from the current main branch:

conda env create -f prodock-env.yml
conda activate prodock
git switch main
git pull --ff-only

Working on New Features

1. Create a branch. Use a short name that describes the feature or fix.

   git switch -c feature/your-feature-name
   

2. Develop and commit. Keep commits focused on one logical change.

   git add <changed-files>
   git commit -m "Describe the change"
   

3. Run the quality checks.

   ./lint.sh
   python -m pytest -q Test
   git diff --check
   

Integrating Changes

Update the feature branch against the latest main, resolve conflicts locally, and repeat the checks:

git fetch origin
git rebase origin/main
python -m pytest -q Test

Push the feature branch and open a pull request targeting main:

git push --set-upstream origin feature/your-feature-name

Important Notes

• Do not commit generated campaign outputs, caches, built documentation, or local environment paths.
• Keep reproduction inputs and scripts visible in the commit tree; see REPRODUCIBILITY.md for the artifact boundary.
• Use pull requests for changes to main and include the validation commands in the pull-request description.

Publication

ProDock: From multi-target consensus docking into database-backed storage

License

This project is licensed under MIT License - see the License file for details.

Authors & Contributors

• Lai Hoang Son Le
• Thanh-An Pham
• Tieu-Long Phan

Acknowledgments

This work has received support from the Korea International Cooperation Agency (KOICA) under the project entitled "Education and Research Capacity Building Project at University of Medicine and Pharmacy at Ho Chi Minh City", conducted from 2024 to 2025 (Project No. 2021-00020-3). TLP and PFS have received funding from the European Union's Horizon Europe Doctoral Network programme under the Marie Skłodowska-Curie grant agreement No. 101072930 (TACsy - Training Alliance for Computational systems chemistry).