AffiniPy is an automated, end-to-end molecular docking and scoring pipeline built in Python.
It integrates RDKit, Dimorphite-DL, Meeko, pdb2pqr, MDAnalysis and a bundled AutoDock Vina binary to streamline reproducible protein–ligand virtual screening, descriptor profiling, and composite scoring.
AffiniPy automates everything from file preparation to ranked ligand output — ideal for rapid structure-based prioritization in drug discovery workflows.
- ⚙️ Automated protein and ligand preparation with unified charge and atom typing models
- 🧪 Protonation with Dimorphite-DL
- 🚀 Docking via AutoDock Vina (parallelized using Joblib)
- 🧮 Descriptor profiling using RDKit (Lipinski, Veber, SA, LE)
- 🧩 Composite affinity scoring and ranking
- 📊 Color-coded Excel summary of results
- 💾 Automated best-pose extraction and receptor–ligand complex generation
- 🧯 Transparent logs and error tracking
Clone this repository and create the environment using Conda.
git clone https://github.com/FreaX997/AffiniPy.git
cd AffiniPy
conda env create -f Informatics.yml
conda activate InformaticsVerify that the required external tools are available:
obabel -V
pdb2pqr --version🧠 Tested on Linux (Ubuntu 22.10), Python 3.11, RDKit 2025.03, and AutoDock Vina 1.2.7. Windows compatibility not tested yet.
To get started, keep all required input files in a single folder.
This folder should include:
AffiniPy/
├─ bin/
│ └─ vina
├── AffiniPy.ipynb
├── sascorer.py
├── fpscores.pkl.gz # companion file for sascorer
├── Informatics.yml
├── 7JGW.pdb # Protein PDB(s)
├── ligand1.sdf # Ligand SDF(s)
├── ligand2.sdf
└── ...
When you run AffiniPy.ipynb, the pipeline automatically creates the following filesystem:
AffiniPy/
├── Data/
│ ├── Protein/
│ └── Ligands_Raw/
├── Prep/
| ├── Ligands_prepared
| └── Proteins_prepared
├── Docking/
│ ├── Result_PDBQTs/
│ ├── BestPoses/
│ └── ...
You’ll only need to:
- Move your protein
.pdbfiles into/Data/Protein/ - Move your ligand
.sdffiles into/Data/Ligands_Raw/
Everything else — preparation, docking, descriptor calculation, and scoring — is fully automated.
- The pipeline expects two protein files:
7JGW.pdb: original structure downloaded from RCSB PDB.7JGW_2.pdb: prepared in UCSF Chimera’s Dock Prep to fix alternate residues, incomplete side chains, and remove water/ligands.
- AffiniPy automatically isolates the co-crystallized ligand if the ligand ID from the RCSB PDB page of the protein is manually specified in Code Cell 3 of the notebook.
In Code Cell 5, define the docking grid box:
Manual mode (recommended for reproducibility):
gbox_centre = [-2.96, -2.55, 26.75]
gbox_size = [11.25, 20.10, 24.24]Automatic mode (optional) (uncomment these lines from Code Cell 5):
gbox_centre = lig_only.center_of_geometry().tolist()
gbox_size = lig_only.positions.max(axis=0) - lig_only.positions.min(axis=0) + 10.0(Requires the co-crystallized ligand to be present.)
- Place multiple
.sdffiles inData/Ligands_Raw/— one ligand per file. - AffiniPy automatically performs:
- Protonation (Dimorphite-DL at pH 7.4)
- 3D conformer generation (RDKit ETKDGv3)
- MMFF94 energy minimization
- Conversion to
.pdbqtusing Meeko
❗ Failed ligands are automatically skipped and logged (e.g., 3D generation or protonation failure).
Open the Jupyter Notebook (AffiniPy.ipynb) in VS Code or JupyterLab and execute cells sequentially.
Each code block corresponds to a distinct stage of the workflow:
Protein Prep → Ligand Prep → Docking → Descriptor Calculation → Composite Scoring → Output
You can re-run from any stage; prepared files are reused automatically.
The output structure after execution:
AffiniPy/
├── Docking/
│ ├── Result_PDBQTs/ # Docking output and logs
│ ├── BestPoses/ # Complexed protein–ligand best poses
│ └── 7JGW_screening_results_summary.xlsx
| Sheet | When It Appears | Description |
|---|---|---|
| Docking + Descriptors | Always | Main results sheet: lists all ligands with docking scores, ligand efficiency, synthetic accessibility (SA), Lipinski/Veber flags, and composite scores. |
| Errors | If some ligands fail during docking | Contains ligand names and specific error messages captured from Vina output. |
| Pose Extraction Errors | If best-pose generation fails | Lists ligands where PDBQT → PDB conversion or protein merging failed. |
AffiniPy ranks ligands by a Composite Score integrating binding energy, efficiency, and synthetic feasibility.
| Parameter | Weight | Direction |
|---|---|---|
| Docking Score | 0.45 | Lower is better |
| Ligand Efficiency | 0.30 | Higher is better |
| Synthetic Accessibility | 0.15 | Lower is better |
| Lipinski / Veber Bonus | +0.05 each | Pass increases score |
💡 The composite score highlights ligands that balance strong binding, efficiency, and synthetic feasibility.
Users can manually adjust the weights in Code Cell 11 to re-prioritize parameters.
An example dataset derived from B. Hassan Ganesh, B. Aruchamy, S. Mudradi, S. Mohanty, H. Padinjarathil, S. Carradori, P. Ramani, ChemMedChem 2025, 20, e202400371.
https://doi.org/10.1002/cmdc.202400371
is included under Example_Data/ to demonstrate input and output formats. To use these example files, please move the files from the Example_Data/ folder into the folder where AffiniPy.ipynb, sascorer.py and fpscores.pkl.gz are located.
All dependencies are listed in Informatics.yml.
Key libraries and tools include:
| Category | Tools |
|---|---|
| Core | Python 3.11+, RDKit, MDAnalysis, pandas, scikit-learn |
| Docking | AutoDock Vina, Meeko |
| Protonation | Dimorphite-DL |
| Descriptor | sascorer |
| Utilities | openbabel, pdb2pqr, joblib, numpy, openpyxl |
🧩 AutoDock Vina is shipped as a precompiled binary inside the /bin folder for portable, reproducible runs. No external installation required.
This project is licensed under the Apache License 2.0 © 2025 Sarthak Mohanty
You are free to use, modify, and distribute this code under the terms of the Apache 2.0 license with proper attribution.
If you use AffiniPy in your research, please cite:
Sarthak Mohanty (2025). AffiniPy: Automated Molecular Docking and Scoring Pipeline.
GitHub: https://github.com/FreaX997/AffiniPy
- Protein structures are auto-prepared via pdb2pqr + Meeko, but unusual cofactors or metals may still require manual curation.
- Grid box must currently be set manually for best reliability
- Composite scoring weights are static (can be changed manually in code)
- Designed and tested on Linux systems; Windows compatibility untested
AffiniPy is a continuously evolving open-source project.
Feature additions, optimization strategies, and modular improvements will be introduced over time as the workflow matures and community feedback grows.
Stay tuned on the repository for updates. 🚀
🦉 Built by Sarthak with curiosity, late-nights, impostor syndrome, and a healthy disregard for un-automated workflows.
“Reproducible docking doesn’t have to be painful.”