OpenFold3 availability on SageMaker JumpStart
AWS has confirmed the availability of OpenFold3 on Amazon SageMaker JumpStart, expanding the catalog of foundation models for biomedical research. This model, developed by the OpenFold Consortium and the AlQuraishi Lab at Columbia University, allows for the prediction of all-atom biomolecular complex structures.
Integration into JumpStart allows customers to deploy the model with just a few clicks from the SageMaker console or via the Python SDK. This availability joins other recent models such as granite-speech-4.1-2b and kanana-2-30b-a3b-instruct, reinforcing AWS's offering for specialized AI use cases.
AWS has confirmed that OpenFold3 is now available on Amazon SageMaker JumpStart, expanding the catalog of foundation models for biomedical research. This model, developed by the OpenFold Consortium and the AlQuraishi Lab at Columbia University, enables the prediction of all‑atom biomolecular complex structures. The integration into JumpStart allows customers to deploy the model with just a few clicks from the SageMaker console or via the Python SDK, streamlining access for researchers and developers who need advanced structure‑prediction capabilities without managing infrastructure.
The availability of OpenFold3 on JumpStart joins other recent models such as granite‑speech‑4.1‑2b and kanana‑2‑30b‑a3b‑instruct, reinforcing AWS’s offering for specialized AI use cases. By providing a ready‑to‑use deployment path, the platform lowers the barrier to entry for teams that require high‑performance diffusion models for protein–protein, protein–DNA, or protein–RNA complex prediction. Users can immediately start experimenting with the model’s multi‑chain and heterogeneous interaction support, while benefiting from SageMaker’s managed scaling and monitoring features.
Structural prediction capabilities
OpenFold3 is a diffusion-based model that extends structure prediction beyond individual proteins. It can model multi-chain complexes and heterogeneous interactions between biomolecules, including proteins, DNA, RNA, and small-molecule ligands.
This capability is relevant for computer-aided drug design and for academic and pharmaceutical research laboratories. Unlike previous models that focused on monomers, OpenFold3 addresses the complexity of complete biomolecular complexes.
Types of biomolecules supported by OpenFold3
The model is designed to predict structures of the following types of biomolecules and complexes.
| Biomolecule type | Description |
|---|---|
| Proteins | Structures of individual polypeptide chains or in complexes. |
| DNA and RNA | Structures of nucleic acids and their interactions with other molecules. |
| Ligands | Small molecules that bind to proteins or nucleic acids. |
| Heterogeneous complexes | Interactions between multiple types of biomolecules simultaneously. |
Inference architecture and performance
Deployment on SageMaker JumpStart uses asynchronous inference and multi-GPU parallelism to handle the computational load required by complex structure prediction. This architecture allows processing high-volume requests without blocking the user interface.
The ability to parallelize across multiple GPUs is crucial for large-scale diffusion models, as it reduces inference time and allows scaling processing capacity according to the needs of the laboratory or enterprise.
Training data and custom alignments
OpenFold3 was trained on datasets that include 245,000 structures from the Protein Data Bank (PDB) and millions of monomers from the MGNIFY database. The training data is available under the CC BY 4.0 license in the AWS Open Data Registry.
A key feature is the option to provide custom multiple sequence alignments (MSAs). This allows users to improve prediction accuracy for specific sequences and maintain control over sensitive or proprietary data without needing to share it with third parties.
Limitations and prior verification
Before presenting results based on OpenFold3, teams must verify the exact version of the model available and its access conditions. Current documentation does not specify particular accuracy benchmarks for this version on JumpStart, so it is recommended to compare with previous project results.
It is also necessary to confirm the available AWS regions, compatible instance types, and inference costs, which may vary depending on the complexity of the structure and the duration of the prediction. The privacy of user-provided data must be evaluated according to the AWS account security policies.
Steps to verify the implementation
Recommended process to evaluate the suitability of OpenFold3 in a specific project.
- 01Consult the SageMaker JumpStart documentation to confirm the region and instance type.
- 02Review the license and terms of use for the specific version of OpenFold3.
- 03Run inference tests with sample data to measure response time and cost.
- 04Compare the accuracy of predictions with known experimental structures if available.
Keep exploring
Sources consulted
Corrections and transparency
If you spot incorrect or outdated information, send us a correction with the page and source we should review.
Submit a correction