IntFold: A Controllable Foundation Model for General and Specialized Biomolecular Structure Prediction.

🚀 Quick Start

To quickly get started with IntFold, you can use the following commands:

# Install IntFold from PyPI
pip install intfold
# Run inference with an example YAML file
intfold predict ./examples/5S8I_A.yaml --out_dir ./output

⚙️ Installation

To more complete installation instructions and usage, please refer to the Installation Guide.

🔍 Inference

Prepare Input File: Create a YAML file with your sequences following our input format specification

Run Prediction:

intfold predict your_input.yaml --out_dir ./results

Check Results: Find predicted structures and confidence scores in the output directory, you can also check the section of output format in output documentation.
Optional Optimization: Enable custom kernels for faster inference and reduced memory usage

For comprehensive usage instructions and examples, refer to the Usage Guide.

📊 Benchmarking

To comprehensively evaluate the performance of IntFold, we conducted a rigorous evaluation on FoldBench. We compared IntFold against several leading methods, including Boltz-1,2, Chai-1, Protenix and Alphafold3.

For more details on the benchmarking process and results, please refer to our Technical Report.

🌐 IntFold Server

We highly recommend using the IntFold Server for the most accurate, complete, and convenient biomolecular structure predictions. It requires no installation and provides an intuitive web interface to submit your sequences and visualize results directly in your browser. The server runs the full, optimized, latest IntFold implementation for optimal performance.

📜 Citation

If you use IntFold in your research, please cite our paper:

@misc{theintfoldteam2025intfoldcontrollablefoundationmodel,
      title={IntFold: A Controllable Foundation Model for General and Specialized Biomolecular Structure Prediction}, 
      author={The IntFold Team and Leon Qiao and Wayne Bai and He Yan and Gary Liu and Nova Xi and Xiang Zhang},
      year={2025},
      eprint={2507.02025},
      archivePrefix={arXiv},
      primaryClass={q-bio.BM},
      url={https://arxiv.org/abs/2507.02025}
}

🔗 Acknowledgements

The implementation of fast layernorm operators is inspired by OneFlow and FastFold, following Protenix's usage.
Many components in intfold/openfold/ are adapted from OpenFold, with substantial modifications and improvements by our team (except for the LayerNorm part).
This repository, the implementation of Inference Data Pipeline(Data/Feature Processing and MSA generation tasks) referred to Boltz-1, and modify some codes to adapt to the input of our model.

⚖️ License

The IntFold project, including code and model parameters, is made available under the Apache 2.0 License, it is free for both academic research and commercial use.

📬 Contact Us

If you have any questions or are interested in collaboration, please feel free to contact us at [email protected].