ByteFF (ByteDance Force Field)

Welcome to the repository of ByteFF! This repository hosts the source code for ByteFF paper.

Use ByteFF

We offer a free trial of ByteFF through our API service.

For further details, please reach out to [email protected].

Model architecture

Data preparation

There are three types of data used in training process:

• Partial hessian data, containing coordinates, partial hessians and forcefield parameters for pretraining.
• Torsion data, containing coordinates, energies and torsion_ids.
• Energy force data, containing coordinates, energies and forces.

See scripts/prepare_data/README.md for detailed instructions.

Training

There are three stage of training:

• Pretraining with partial hessian data.
• Traing with partial hessian and torsion data.
• Finetuning with energy force data.

See scripts/train/README.md for detailed instructions.

Output forcefield parameters

The trained model can be used to infer forcefield parameters given a molecule.
The output format is .itp file, which is compatible with Gromacs.

Use scripts/write_itp/write_itp.py and scripts/write_itp/write_itp_ensemble.py

Data availability

BDTorsion benchmark data are accessible in the data folder. Please adhere to the Apache 2.0 license and cite our paper if you utilize this data.

Citing ByteFF

If you use ByteFF in your research, please cite:

@Article{D4SC06640E,
author ="Zheng, Tianze and Wang, Ailun and Han, Xu and Xia, Yu and Xu, Xingyuan and Zhan, Jiawei and Liu, Yu and Chen, Yang and Wang, Zhi and Wu, Xiaojie and Gong, Sheng and Yan, Wen",
title  ="Data-driven parametrization of molecular mechanics force fields for expansive chemical space coverage",
journal  ="Chem. Sci.",
year  ="2025",
pages  ="-",
publisher  ="The Royal Society of Chemistry",
doi  ="10.1039/D4SC06640E",
url  ="http://dx.doi.org/10.1039/D4SC06640E"}

License

This project is licensed under the Apache License, Version 2.0.