Overview#

OpenFold3 is an all-atom biomolecular complex structure prediction model from the OpenFold Consortium and the AlQuraishi Laboratory. The previously developed model Openfold is a PyTorch implementation of the JAX-based AlphaFold2 model reported in Highly accurate protein structure prediction with AlphaFold. OpenFold and AlphaFold2 predict the structure of single-chain proteins. Similarly, OpenFold3 is a pytorch implementation of the JAX-based AlphaFold3 reported in Accurate structure prediction of biomolecular interactions with AlphaFold 3, with inference code at google-deepmind/alphafold3. OpenFold3 and AlphaFold3 extend protein structure prediction capabilities to model complete biomolecular complexes including proteins, DNA, RNA, and small molecule ligands.

The NVIDIA OpenFold3 NIM can:

  • Predict the all-atom 3D structure of biomolecular complexes composed of proteins, DNA, RNA, and ligands.

  • Generate multiple 3D structure predictions, each having a confidence score.

  • Use protein and RNA multiple sequence alignment (MSA) inputs, where the protein MSA inputs can be paired or unpaired.

  • Accept structural templates in CIF format to guide protein structure predictions (available starting from version 1.1.0).

Key Features#

  • Accepts inputs with diverse biomolecular and small-molecule components

    • Proteins: Amino acid sequences

    • DNA: Double-stranded or single-stranded DNA

    • RNA: RNA sequences for structure and interaction modeling

    • Ligands: Small molecules via SMILES strings or CCD codes

  • Predicts the structure of complete biological assemblies including

    • Protein-protein complexes

    • Protein-nucleic acid complexes

    • Protein-ligand complexes

  • Predicts positions of all atoms, not just backbone, providing detailed structural information for analysis.

  • Accepts inputs with protein structural templates (single-chain)

    • Accepts experimental or predicted structures in CIF format

    • Helps improve accuracy when similar structures are available

    • Supports template-guided structure prediction for proteins

Note: The OpenFold3 NIM accepts paired MSA inputs, but does not perform ‘online’ pairing of MSAs.

Tooling#

The OpenFold3 NIM is built with tooling which provides:

  • Fast GPU inference with optimized CUDA kernels

  • Memory-efficient attention mechanisms

  • Support for long sequences and large complexes

  • TensorRT optimization for production deployment

  • NIM Telemetry for faster, more reliable, and privacy-protected performance

Citation:

If you use AlphaFold3, please cite:

@article{Abramson2024,
  author  = {Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans, Richard and Green, Tim and Pritzel, Alexander and Ronneberger, Olaf and Willmore, Lindsay and Ballard, Andrew J. and Bambrick, Joshua and Bodenstein, Sebastian W. and Evans, David A. and Hung, Chia-Chun and O’Neill, Michael and Reiman, David and Tunyasuvunakool, Kathryn and Wu, Zachary and Žemgulytė, Akvilė and Arvaniti, Eirini and Beattie, Charles and Bertolli, Ottavia and Bridgland, Alex and Cherepanov, Alexey and Congreve, Miles and Cowen-Rivers, Alexander I. and Cowie, Andrew and Figurnov, Michael and Fuchs, Fabian B. and Gladman, Hannah and Jain, Rishub and Khan, Yousuf A. and Low, Caroline M. R. and Perlin, Kuba and Potapenko, Anna and Savy, Pascal and Singh, Sukhdeep and Stecula, Adrian and Thillaisundaram, Ashok and Tong, Catherine and Yakneen, Sergei and Zhong, Ellen D. and Zielinski, Michal and Žídek, Augustin and Bapst, Victor and Kohli, Pushmeet and Jaderberg, Max and Hassabis, Demis and Jumper, John M.},
  journal = {Nature},
  title   = {Accurate structure prediction of biomolecular interactions with AlphaFold 3},
  year    = {2024},
  volume  = {630},
  number  = {8016},
  pages   = {493–-500},
  doi     = {10.1038/s41586-024-07487-w}
}

Advantages of NIMs#

NIMs offer a simple and easy-to-deploy route for self-hosted AI applications. Two major advantages that NIMs offer for system administrators and developers are:

  1. Increased productivity: NIMs allow developers to build generative AI applications quickly, in minutes rather than weeks, by providing a standardized way to add AI capabilities to their applications.

  2. Simplified deployment: NIMs provide containers that can be easily deployed on various platforms, including clouds, data centers, or workstations, making it convenient for developers to test and deploy their applications.

  3. Acceleration: The OpenFold3 NIM has a TRT backend with accelerations that are not available open-source. Our measurements, see Structure Prediction Performance, show that with the TRT backend (and cuEquivariance kernels) inference is up to 1.7x faster than the Pytorch + DeepSpeed configuration.

The OpenFold3 NIM provides a fast, accurate model behind a consistent API for predicting biomolecular complex structures. As part of the broader NVIDIA NIM Ecosystem, OpenFold3 can be used in conjunction with other NIMs to build pipelines for drug discovery, protein engineering, and understanding biological mechanisms at the molecular level.

You are responsible for ensuring that your use of NVIDIA AI Foundation Models complies with all applicable laws.