bytedance / Protenix

Protenix: Protein + X > 📣📣📣 **We're hiring!** \ > Positions in **_Beijing_** 🇨🇳 and **_Seattle_** 🇺🇸 \ > Interested in machine learning, computational chemistry/biology, structural biology, or drug discovery? Join us to build cutting-edge AI for biology! \ > 👉 **Join us »** ⚡ Protenix Web Server • 📄 Technical Report We’re excited to introduce **Protenix** — Toward High-Accuracy Open-Source Biomolecular Structure Prediction. Protenix is built for high-accuracy structure prediction. It serves as an initial step in our journey toward advancing accessible and extensible research tools for the computational biology community. 🌟 Related Projects • **PXDesign** is a model suite for de novo protein-binder design built on the Protenix foundation model. PXDesign achieves 20–73% experimental success rates across multiple targets — 2–6× higher than prior SOTA methods such as AlphaProteo and RFdiffusion. The framework is freely accessible via the Protenix Server. • **PXMeter** is an open-source toolkit designed for reproducible evaluation of structure prediction models, released with high-quality benchmark dataset that has been manually reviewed to remove experimental artifacts and non-biological interactions. The associated study presents an in-depth comparative analysis of state-of-the-art models, drawing insights from extensive metric data and detailed case studies. The evaluation of Protenix is based on PXMeter. • **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. 🎉 Latest Updates • **2026-02-05: Protenix-v1 Released** 💪 [Technical Report] • Supported Template/RNA MSA features and improved training dynamics, along with further Inference-time model performance enhancements. • **2025-11-05: Protenix-v0.7.0 Released** 🚀 • Introduced advanced diffusion inference optimizations: Shared variable caching, efficient kernel fusion, and TF32 acceleration. See our performance analysis. • **2025-07-17: Protenix-Mini & Constraint Features** • Released lightweight model variants (Protenix-Mini) that drastically reduce inference costs with minimal accuracy loss. • Added support for atom-level contact and pocket constraints, enhancing prediction accuracy through physical priors. • **2025-01-16: Pipeline Enhancements** • Open-sourced the full training data pipeline and MSA pipeline. • Integrated local ColabFold-compatible search for streamlined MSA generation. 🚀 Getting Started 🛠 Quick Installation 🧬 Quick Prediction Key Model Descriptions | Model Name | MSA | RNA MSA | Template | Params | Training Data Cutoff | Model Release Date | | :--- | :---: | :---: | :---: | :---: | :---: | :---: | | | ✅ | ✅ | ✅ | 368 M | 2021-09-30 | 2026-02-05 | | | ✅ | ✅ | ✅ | 368 M | 2025-06-30 | 2026-02-05 | | | ✅ | ❌ | ❌ | 368 M | 2021-09-30 | 2025-05-30 | • **protenix_base_default_v1.0.0**: Default model, trained with a data cutoff aligned with AlphaFold3 (2021-09-30). > 💡 > This is the **highly recommended** model for conducting fair, rigorous public benchmarks and comparative studies against other state-of-the-art methods. • **protenix_base_20250630_v1.0.0**: Applied model, trained with an updated data cutoff (2025-06-30) for better practical performance. This model can be used for practical application scenarios. • **protenix_base_default_v0.5.0**: Previous version of the model, maintained primarily for backward compatibility with users who developed based on v0.5.0. For a complete list of supported models, please refer to Supported Models. For detailed instructions on installation, data preprocessing, inference, and training, please refer to the Training and Inference Instructions. We recommend users refer to inference_demo.sh for detailed inference methods and input explanations. 📊 Benchmark **Protenix-v1 (refers to the model)**, the first fully open-source model that outperforms AlphaFold3 across diverse benchmark sets while adhering to the same training data cutoff, model scale, and inference budget as AlphaFold3. For challenging targets, such as antigen-antibody complexes, the prediction accuracy of Protenix-v1 can be further enhanced through inference-time scaling – increasing the sampling budget from several to hundreds of candidates leads to consistent log-linear gains. For detailed benchmark metrics on each dataset, please refer to docs/model_1.0.0_benchmark.md. Citing Protenix If you use Protenix in your research, please cite the following: 📚 Citing Related Work Protenix is built upon and inspired by several influential projects. If you use Protenix in your research, we also encourage citing the following foundational works where appropriate: Contributing to Protenix We welcome contributions from the community to help improve Protenix! 📄 Check out the Contributing Guide to get started. ✅ Code Quality: We use hooks to ensure consistency and code quality. Please install them before making commits: 🐞 Found a bug or have a feature request? Open an issue. Acknowledgements The implementation of LayerNorm operators refers to both OneFlow and FastFold. We also adopted several module implementations from OpenFold, except for , which is implemented independently. Code of Conduct We are committed to fostering a welcoming and inclusive environment. Please review our Code of Conduct for guidelines on how to participate respectfully. 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 or vulnerability reporting email. Please do **not** create a public GitHub issue. License The Protenix project including both code and model parameters is released under the Apache 2.0 License. It is free for both academic research and commercial use. Contact Us We welcome inqu…