Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/bytedance/protenix

A trainable PyTorch reproduction of AlphaFold 3.
https://github.com/bytedance/protenix

ai4science research

Last synced: 13 days ago
JSON representation

A trainable PyTorch reproduction of AlphaFold 3.

Awesome Lists containing this project

README 

        
# Protenix: Protein + X



A trainable PyTorch reproduction of [AlphaFold 3](https://www.nature.com/articles/s41586-024-07487-w).



For more information on the model's performance and capabilities, see our technical report ([biorxiv](https://www.biorxiv.org/content/10.1101/2025.01.08.631967v1) | [pdf](Protenix_Technical_Report.pdf)).



You can follow our [twitter](https://x.com/ai4s_protenix) or join the conversation in the [discord server](https://discord.gg/8ZMWy89aMf).



![Protenix predictions](assets/protenix_predictions.gif)



**🌟🌟🌟  We have also open sourced [Protenix-Dock](https://github.com/bytedance/Protenix-Dock), our implementation of a classical protein-ligand docking framework that leverages empirical scoring functions. Without using deep neural networks, Protenix-Dock delivers competitive performance in rigid docking tasks.** 



## ⚡ Try it online

- [Web server link](https://protenix-server.com)



## 🔥 Feature Update

* 🚀 The preview version of [constraint feature](./README.md#early-access-to-new-constraint-feature) is released to branch [`constraint_esm`](https://github.com/bytedance/Protenix/tree/constraint_esm).

* 🪐 The [training data pipeline](./docs/prepare_training_data.md) is released.

* ⚡️  The [MSA pipeline](./docs/msa_pipeline.md) is released.

* 🛸 Use [local colabfold_search](./docs/colabfold_compatiable_msa.md) to generate protenix-compatible MSA.



## Installation



### Run with PyPI (recommended):



```bash

pip3 install protenix

```

### Run with Docker:



If you're interested in model training, we recommand to [ run with docker](docs/docker_installation.md).



### Local installation (cpu only)

For development on a CPU-only machine, it is convenient to install with the `--cpu` flag in editable mode:

```

python3 setup.py develop --cpu

```



## Inference



### Command line inference



If you set up `Protenix` by `pip`, you can run the following command to do model inference:



```bash

# the default n_cycle/n_step/n_samples is 10/200/5 respectively, you can modify it by passing --cycle x1 --step x2 --sample x3



# run with example.json, which contains precomputed msa dir.

protenix predict --input examples/example.json --out_dir  ./output --seeds 101



# run with multiple json files, the default seed is 101.

protenix predict --input ./jsons_dir/ --out_dir  ./output



# if the json do not contain precomputed msa dir,

# add --use_msa_server to search msa and then predict.

# if mutiple seeds are provided, split them by comma.

protenix predict --input examples/example_without_msa.json --out_dir ./output --seeds 101,102 --use_msa_server

```



### Convert PDB/CIF file to json



If your input is pdb or cif file, you can convert it to json file for inference.



```bash

# ensure `release_data/ccd_cache/components.cif` or run:

python scripts/gen_ccd_cache.py -c release_data/ccd_cache/ -n [num_cpu]



# for PDB

# download pdb file

wget https://files.rcsb.org/download/7pzb.pdb

# run with pdb/cif file, and convert it to json file for inference.

protenix tojson --input examples/7pzb.pdb --out_dir ./output



# for CIF (same process)

# download cif file

wget https://files.rcsb.org/download/7pzb.cif

# run with pdb/cif file, and convert it to json file for inference.

protenix tojson --input examples/7pzb.cif --out_dir ./output

```



### Performance details



**Detailed information on the format of the input JSON file and the output files can be found in [ input and output documentation ](docs/infer_json_format.md)**.



Alternatively you can run inference by:



Note: by default, we do not use layernorm and EvoformerAttention kernels for simple configuration, if you want to speed up inference, see [ setting up kernels documentation ](docs/kernels.md).



```bash

bash inference_demo.sh

```



Arguments in this scripts are explained as follows:



* `input_json_path`: path to a JSON file that fully describes the input.

* `dump_dir`: path to a directory where the results of the inference will be saved.

* `dtype`: data type used in inference. Valid options include `"bf16"` and `"fp32"`.

* `use_msa`: whether to use the MSA feature, the default is true.

* `use_esm`: whether to use the ESM feature, the default is false.



### Convert PDB/CIF file to json



If your input is pdb or cif file, you can convert it to json file for inference.

```bash

# run with pdb/cif file, and convert it to json file for inference.

protenix tojson --input examples/7pzb.pdb --out_dir ./output

```



### MSA search

We also provide an independent MSA search function, you can do msa search from json file or fasta file.

```bash

# run msa search with json file, it will write precomputed msa dir info to a new json file.

protenix msa --input examples/example_without_msa.json --out_dir ./output



# run msa search with fasta file which only contains protein.

protenix msa --input examples/prot.fasta --out_dir ./output

```



### Run with PyMol



If you want to run Protenix inference with `PyMol`, please refer to [PyMOLfold](https://github.com/colbyford/PyMOLfold).



## Training

If you're interested in model training, see [ training documentation ](docs/training.md).



## Performance

#### **Model Performance across Several Benchmarks**

![Overall Metrics](assets/overall_metrics.png)



#### ***Early Access to NEW Constraint Feature!***



🎉 Protenix now allows users to specify ***contacts***, enabling the model to leverage additional inter-chain information as constraint guidance! We benchmarked our constraint feature on Posebuster and a protein-antibody interfaces subset. Protenix demonstrates powerful ability in predicting more accurate structures. If you want to have a try, checkout to branch `constraint_esm` for details about the input format.



![Constraint Metrics](assets/constraint_metrics.png)



> **Tips:** Our online service already supports the new features, so feel free to try it now! Due to the preview version, the constraint support is only applicable in the branch `constraint_esm`. If you want to run inference via the command line, please check out to this branch first.



## Training and Inference Cost



See the [model_train_inference_cost documentation](docs/model_train_inference_cost.md) for memory and time consumption in training and inference.



## Citing This Work



If you use this code or the model in your research, please cite the following paper:



```

@article{chen2025protenix,

  title={Protenix - Advancing Structure Prediction Through a Comprehensive AlphaFold3 Reproduction},

  author={Chen, Xinshi and Zhang, Yuxuan and Lu, Chan and Ma, Wenzhi and Guan, Jiaqi and Gong, Chengyue and Yang, Jincai and Zhang, Hanyu and Zhang, Ke and Wu, Shenghao and Zhou, Kuangqi and Yang, Yanping and Liu, Zhenyu and Wang, Lan and Shi, Bo and Shi, Shaochen and Xiao, Wenzhi},

  year={2025},

  doi = {10.1101/2025.01.08.631967},

  journal = {bioRxiv}

}

```



## Acknowledgements



Implementation of the layernorm operators referred to [OneFlow](https://github.com/Oneflow-Inc/oneflow) and [FastFold](https://github.com/hpcaitech/FastFold). We used [OpenFold](https://github.com/aqlaboratory/openfold) for some [module](protenix/openfold_local/) implementations, except the [`LayerNorm`](protenix/model/layer_norm/).



## Contribution



Please check [Contributing](CONTRIBUTING.md) for more details. If you encounter problems using Protenix, feel free to create an issue! We also welcome pull requests from the community.



```bash

pip install pre-commit

pre-commit install

```



So new commits will be automatically checked.



## Code of Conduct



Please check [Code of Conduct](CODE_OF_CONDUCT.md) for more details.



## Security



If you discover a potential security issue in this project, or think you may

have discovered a security issue, we ask that you notify Bytedance Security via our [security center](https://security.bytedance.com/src) or [vulnerability reporting email]([email protected]).



Please do **not** create a public GitHub issue.



## License



The Protenix project, including code and model parameters, is made available under the [Apache 2.0 License](./LICENSE), it is free for both academic research and commercial use.



We welcome inquiries and collaboration opportunities for advanced applications of our model, such as developing new features, fine-tuning for specific use cases, and more. Please feel free to contact us at [email protected].