VinaSF PyTorch

This repository provides a PyTorch-friendly version of the Vina scoring function. It is a fork of the excellent guyuehuo/opendock project.

Installation

Install from Git (recommended)

You can install the latest version directly from GitHub with pip:

pip install git+https://github.com/kim-iljung/VinaSF_PyTorch.git

This command uses the pyproject.toml configuration in the repository so no additional flags (such as #subdirectory=) are needed.

Local development install

If you have cloned the repository locally, install it in editable mode to pick up changes without reinstalling:

pip install -e .

Usage example

The snippet below shows how to evaluate the Vina scoring function on a receptor/ligand pair using RDKit and PyTorch. The same pattern works for coordinates you generate elsewhere.

from rdkit import Chem
import torch

from vinasf_torch import VinaSFTorch

receptor = Chem.MolFromPDBFile("receptor.pdb", removeHs=False)
ligand = Chem.SDMolSupplier("ligand.sdf", removeHs=False)[0]

vina = VinaSFTorch.from_rdkit(receptor, ligand)

# Use either the current ligand coordinates or supply your own tensor with
# shape (N, A, 3) for N poses and A heavy atoms.
coords = vina.ligand.pose_heavy_atoms_coords.clone()
score, gradient = vina.score_and_gradient(coords)

print("Score per pose:", score)
print("Gradient tensor:", gradient)

Calling .to(device) on VinaSFTorch moves both the module and any cached receptor/ligand tensors to the requested device, so the entire score evaluation can run on GPU without manual tensor transfers.

Contributing

Contributions, bug reports, and feature requests are welcome! Please open an issue or submit a pull request so that improvements can be discussed.

License

This project is distributed under the terms of the MIT License.