README.md

---
pipeline_tag: other
---

# DyneTrion: Spatiotemporally Coherent Generative Emulation of Protein Dynamics Across Timescales

This repository contains the pretrained weights for **DyneTrion**, a generative protein dynamics emulator that jointly enforces geometric symmetry, structural consistency, and temporal coherence within a single framework.

- **Paper:** [DyneTrion: A Spatio-temporally Coherent Generative Emulator for Protein Dynamics Across Timescales](https://huggingface.co/papers/2607.15309)
- **GitHub Repository:** [fudan-generative-vision/DyneTrion](https://github.com/fudan-generative-vision/DyneTrion)

<table class="center">
  <tr>
    <td style="text-align: center"><b>1ail_A</b></td>
    <td style="text-align: center"><b>1ifg_A</b></td>
    <td style="text-align: center"><b>2kxl_A</b></td>
    <td style="text-align: center"><b>2rcs_H</b></td>
  </tr>
  <tr>
    <td style="text-align: center">
      <img src="https://raw.githubusercontent.com/fudan-generative-vision/DyneTrion/main/assets/1ail_A.gif" width="100%">
    </td>
    <td style="text-align: center">
      <img src="https://raw.githubusercontent.com/fudan-generative-vision/DyneTrion/main/assets/1ifg_A.gif" width="100%">
    </td>
    <td style="text-align: center">
      <img src="https://raw.githubusercontent.com/fudan-generative-vision/DyneTrion/main/assets/2kxl_A.gif" width="100%">
    </td>
    <td style="text-align: center">
      <img src="https://raw.githubusercontent.com/fudan-generative-vision/DyneTrion/main/assets/2rcs_H.gif" width="100%">
    </td>
  </tr>
</table>

## Installation

```bash
# Create virtual environment (Python 3.10.12 is recommended)
python -m venv .venv
source .venv/bin/activate

# Install PyTorch (CUDA 12.4)
pip install torch==2.4.0 --index-url https://download.pytorch.org/whl/cu124

# Install other dependencies
pip install -r requirements.txt
```

## Inference

Run inference using:

```bash
bash inference.sh
```

- Model checkpoint: `step_400000.pth`
- Input CSV: `datasets/inference/inference_data.csv`
- Frame number: `n_frame = 16`
- Motion number: `n_motion = 2`
- Frame sampling step: `sample_step = 40`
- Extrapolation time: `extrapolation_time = 16`
- Noise scale: `noise_scale = 1.0`

Inference results will be saved to `save_root` (default: `./test/inference/`).

For more details, please check the [GitHub Repository](https://github.com/fudan-generative-vision/DyneTrion).

## Citation

If you find DyneTrion useful for your research, please cite the paper:

```bibtex
@article{cheng2025dynetrion,
  title={DyneTrion: A Spatio-temporally Coherent Generative Emulator for Protein Dynamics Across Timescales},
  author={Cheng, Kaihui and Cai, Zhiqiang and Tu, Peng and Yao, Yisong and Han, Limei and Wu, Libo and Zhu, Siyu and Yang, Tzuhsiung and Qi, Yuan},
  journal={arXiv preprint arXiv:2607.15309},
  year={2026}
}
```