Bindigo

Protein-Ligand Binding Affinity Prediction using Molecular Docking and Machine Learning

Bindigo is a Python package for predicting protein-ligand binding affinities. It combines molecular docking (AutoDock Vina) with machine learning to provide fast, accurate predictions for early-stage drug discovery.

Status: v0.1.0 is an alpha skeleton. The CLI, input validation, and dependency checks work. The docking and machine learning steps are not implemented yet, so bindigo predict does not yet produce a binding affinity or write a results file. The output shown below is the target behaviour, not what this release does.

Features

• Hybrid Prediction: Combines AutoDock Vina docking with pre-trained ML models
• Simple Interface: A single bindigo command-line tool
• Automated Workflow: Handles all preprocessing automatically
• Database Integration: Fetch structures from PDB, compounds from ChEMBL
• Flexible Input: Accepts PDB IDs, files, SMILES strings, or SDF files
• Fast Results: Get predictions in minutes

Installation

From PyPI (when released)

pip install bindigo

From Source (Development)

git clone https://github.com/Siavashghaffari/Bindigo.git
cd Bindigo
pip install -e .

Dependencies

Bindigo requires Python 3.9+ and the following packages:

• RDKit (cheminformatics)
• BioPython (protein handling)
• AutoDock Vina (molecular docking)
• scikit-learn (machine learning)
• Click (CLI interface)

All of these except AutoDock Vina are installed automatically via pip.

AutoDock Vina must be installed separately - it is a native binary and pip cannot install it:

conda install -c conda-forge vina      # conda
brew install autodock-vina             # macOS

On Linux and Windows, download a binary from the AutoDock Vina releases page and add it to your PATH. If Vina is installed somewhere else, point Bindigo at it with the BINDIGO_VINA_PATH environment variable:

export BINDIGO_VINA_PATH=/full/path/to/vina

Bindigo checks for Vina before running and prints installation instructions if it is missing.

Quick Start

Command Line Interface

Basic prediction using PDB ID and SMILES:

bindigo predict --protein 1HSG --ligand "CC(=O)Nc1ccc(O)cc1" --output results.csv

Using local files:

bindigo predict --protein protein.pdb --ligand ligand.sdf --output results.csv

Custom binding site:

bindigo predict --protein 1HSG --ligand "CCO" --center 10.5 20.3 15.2 --size 25 --output results.csv

Verbose output:

bindigo predict --protein 1HSG --ligand "CCO" --output results.csv --verbose

Example Output (planned)

The following shows the intended output once docking and ML prediction are implemented; v0.1.0 does not produce it yet.

╔══════════════════════════════════════════════════════════════════╗
║                  Bindigo v0.1.0                                  ║
║         Protein-Ligand Binding Affinity Prediction              ║
╚══════════════════════════════════════════════════════════════════╝

[1/5] Loading inputs...
  ✓ Ligand: CC(=O)Nc1ccc(O)cc1 (Acetaminophen)
  ✓ Protein: Fetching PDB ID 1HSG from RCSB...

[2/5] Preparing protein...
  ✓ Loaded structure: HIV-1 Protease (198 residues, Chain A)

[3/5] Preparing ligand...
  ✓ Generated 3D structure (11 atoms)

[4/5] Running molecular docking...
  ✓ Best pose: -6.2 kcal/mol

[5/5] Predicting binding affinity...
  ✓ ML model prediction complete

╔══════════════════════════════════════════════════════════════════╗
║                      PREDICTION RESULTS                          ║
╠══════════════════════════════════════════════════════════════════╣
║  Predicted Kd:           245.7 nM                                ║
║  Confidence:             High                                    ║
║  Docking Score:          -6.2 kcal/mol                           ║
╚══════════════════════════════════════════════════════════════════╝

Results saved to: results.csv
✓ Prediction completed in 2m 34s

Output Format (planned)

CSV Output

ligand_id,smiles,predicted_kd_nM,confidence,docking_score_kcal_mol,pose_file
ligand_1,CC(=O)Nc1ccc(O)cc1,245.7,High,-6.2,ligand_1_pose.pdb

Fields:

• predicted_kd_nM: Predicted dissociation constant in nanomolar
• confidence: High/Medium/Low based on model applicability
• docking_score_kcal_mol: AutoDock Vina docking score
• pose_file: Path to docked structure PDB file

Documentation

• Installation Guide: See above
• User Guide: Coming soon
• API Reference: Coming soon
• Examples: See examples/ directory

Use Cases

• Virtual Screening: Screen compound libraries against protein targets
• Lead Optimization: Evaluate binding affinities of compound analogs
• Drug Repurposing: Test existing drugs against new targets
• Research: Computational binding affinity studies

Limitations

Current version (v0.1.0) has the following limitations:

• Single predictions only (no batch mode yet)
• Rigid protein docking (no flexible residues)
• Small molecule ligands only (MW < 1000 Da)
• CPU only (no GPU acceleration)

License

MIT - see LICENSE.

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

Citation

If you use Bindigo in your research, please cite:

Bindigo: A Python package for protein-ligand binding affinity prediction
https://github.com/Siavashghaffari/Bindigo

Bindigo uses the following tools:

• AutoDock Vina: Eberhardt et al., J. Chem. Inf. Model. 2021
• RDKit: https://www.rdkit.org
• BioPython: Cock et al., Bioinformatics 2009
• scikit-learn: Pedregosa et al., JMLR 2011

Acknowledgments

• PDBbind database for training data
• RCSB PDB for protein structures
• RDKit and BioPython communities

Development Status

Current version: 0.1.0 (Alpha)

This is an early development version. APIs may change. Production use is not recommended yet.

Authors

This work was developed by Siavash Ghaffari. For any questions, feedback, or additional information, please feel free to reach out. Your input is highly valued and will help improve and refine this pipeline further.