Boltz-1 is an innovative, open-source biomolecular interaction model designed to democratize the prediction of three-dimensional structures of biomolecules. With Boltz-1, researchers can easily predict the structures of proteins, RNA, DNA, and small molecules, while accommodating for modified residues, covalent ligands, and glycans. Furthermore, the model allows users to condition the generation of structures based on pocket residues, enhancing accuracy and specificity in modeling.

For a deeper understanding of the underlying methodologies and capabilities of Boltz-1, please refer to our technical report.

Key Features

• Versatile Inference: Boltz-1 supports multiple input formats for predictions, including:
  • Fasta files for standard uses.
  • A comprehensive YAML schema for advanced requirements.
  • A directory containing files for batch processing.
    Run inference using the command:

boltz predict input_path

To explore all available options, simply use boltz predict --help.

Training and Development

Whether you're interested in retraining the model or contributing to its development, we invite you to join us. You can find detailed guidance on retraining in our training instructions.

Community and Contributions

Contributions are vital for the evolution of Boltz-1, and we are keen to engage with the community. Join us on our Slack channel to discuss new developments, share ideas, and collaborate on projects surrounding Boltz-1.

Future Enhancements

Boltz-1 is continually evolving! Upcoming features include:

• Pocket conditioning support
• Expanded examples
• Comprehensive data processing pipeline
• Colab notebook for inference
• Confidence model checkpoint
• Support for custom paired MSA
• Kernel integration

With Boltz-1, elevate your biomolecular modeling endeavors and harness the full potential of structural predictions.