Boltzina

Boltzina is a pipeline that combines AutoDock Vina docking with Boltz-2 scoring for molecular docking and affinity prediction.

Installation

# Using uv (recommended)
uv venv
uv sync

# Or using pip
pip install .

Related environments including AutoDock Vina, Maxit, and Boltz-2 model checkpoint files can be installed with the following command:

./setup.sh

Modes

Boltzina supports two operation modes:

1. Full docking mode: Performs AutoDock Vina docking followed by Boltz-2 scoring
2. Scoring-only mode: Scores pre-existing ligand poses using only Boltz-2 (no docking)

Usage

Run the pipeline using a configuration file:

python run.py sample/CDK2/config.json

Example for scoring mode:

python run.py sample/CDK2/config_scoring.json

Ligand File Format

To generate a Boltzina-compatible input PDB file and a mols_dict pkl from SMILES, follow the steps below:

# Prepare ligand PDB Files
$ python ligand_preparation.py preparation_sample/input_smiles.txt --output_dir preparation_sample
# Run Boltzina
$ python run.py preparation_sample/config.json

For input file formats, please refer to INPUT_FORMAT.md.

Command Line Options

• config: Path to configuration JSON file (required)
• --batch_size: Batch size for Boltz-2 scoring (default: 1, strongly recommended)
• --num_workers: Number of workers for AutoDock Vina (default: 1)
• --vina_override: Override existing AutoDock Vina results
• --boltz_override: Override existing Boltz-2 scoring results

Configuration File Format

The configuration file should be a JSON file with the following required fields:

{
    "work_dir": "sample/CDK2/boltz_results_base",
    "vina_config": "sample/CDK2/input.txt",
    "fname": "1ckp_cdk2",
    "input_ligand_name": "UNL",
    "output_dir": "sample/CDK2/results",
    "receptor_pdb": "sample/CDK2/boltz_results_base/predictions/1ckp_cdk2/1ckp_cdk2_model_0_protein.pdb",
    "ligand_files": [
        "sample/CDK2/input_pdbs/CDK2_active_0.pdb",
        "sample/CDK2/input_pdbs/CDK2_active_1.pdb"
    ]
}

Before running boltzina, you need to run Boltz-2 to create manifest.json and constraints. For example, run the following:

boltz predict sample/CDK2/base.yaml --out_dir sample/CDK2 --use_msa_server

You need to specify the output directory sample/CDK2/boltz_results_base as work_dir.

Configuration Parameters

• work_dir: You need to specify the output directory of Boltz-2.
• vina_config: Path to AutoDock Vina configuration file (contains binding site coordinates)
• fname: Base filename for output files
• input_ligand_name: Name of the ligand in the input files
• output_dir: Directory where final results will be saved
• receptor_pdb: Path to the receptor PDB file
• ligand_files: Array of paths to ligand files (Now only supports PDB format)

Reference

Furui, K, & Ohue, M. Boltzina: Efficient and Accurate Virtual Screening via Docking-Guided Binding Prediction with Boltz-2. AI for Accelerated Materials Design - NeurIPS 2025. https://openreview.net/forum?id=OwtEQsd2hN