Multi-Modal Foundation · OpenFold
OpenFold3
Predicts 3D structures of proteins, nucleic acids, ligands, and complexes for structural biology and drug discovery.
What OpenFold3 is
OpenFold3 is a biomolecular structure prediction model created by the AlQuraishi Lab at Columbia University and the OpenFold consortium. It is designed as an open reproduction of DeepMind's AlphaFold3, aimed at predicting the three-dimensional structures of standard and non-canonical protein, RNA, and DNA chains, as well as small-molecule ligands and their assemblies.
The model addresses the need for an open-source, community-supported tool capable of multi-modal biomolecular modelling. By offering structure predictions for proteins alongside nucleic acids and chemical ligands, it supports applications across computational chemistry, structural biology, and drug discovery research.
How it works
OpenFold3 replicates the input features described in the AlphaFold3 publication. For sequence analysis, multiple sequence alignments (MSAs) can be generated either through the ColabFold server or by using JackHMMER and hhblits. In addition, the model can incorporate structure templates for protein monomers.
Inference is accelerated using custom computational kernels from cuEquivariance and DeepSpeed4Science. The implementation supports multi-query workflows, distributed execution across multiple GPUs, and low-memory operating modes designed for resource-constrained hardware.
Key facts
- Category
- Protein Structure Prediction
- Supported modalities
- Standard and non-canonical proteins, RNA, DNA, and small molecules
- MSA tools
- ColabFold server, JackHMMER, hhblits
- Kernel acceleration
- cuEquivariance and DeepSpeed4Science
What you can use it for
- Protein and RNA Monomer Modelling
- Predict three-dimensional structures of single protein and RNA chains, including standard and non-canonical residues.
- Protein-Ligand Complex Prediction
- Model binding interactions and complex conformations between proteins and small-molecule chemical ligands for drug discovery.
- Protein-Protein Complex Assembly
- Predict the quaternary structures and interfaces of multi-chain protein assemblies.
- Nucleic Acid Complex Modelling
- Predict structures involving combinations of DNA, RNA, proteins, and chemical compounds.
Run it here
We list this model but cannot run it yet: gated on the Hub — never run past a gate the user did not accept.
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Matches AlphaFold3 performance benchmarks on monomeric RNA structures.
- Provides support for non-canonical residues in protein, RNA, and DNA chains as well as small molecules.
- Includes GPU kernel acceleration via cuEquivariance and DeepSpeed4Science alongside dedicated settings for memory-constrained hardware.
- Supports multi-query job distribution across multiple GPUs.
Watch out for
- This preview release has not yet achieved full parity across all modalities with AlphaFold3.
- Training workflows and datasets for training on custom non-PDB data are still under active development and not yet released.
- Structure template inputs are restricted to protein monomers.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · apache-2.0
This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
Questions
Can I use OpenFold3 commercially?
Check the licence before commercial use. This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
What biomolecular input types does OpenFold3 support?
It supports standard and non-canonical protein, RNA, and DNA chains, as well as small-molecule ligands.
How are multiple sequence alignments generated for OpenFold3?
MSAs can be generated using the ColabFold server pipeline or via JackHMMER and hhblits following the AlphaFold3 protocol.
Can OpenFold3 run on hardware with limited GPU memory?
Yes, it contains custom configuration options specifically designed for memory-constrained GPU environments.
Does OpenFold3 support structural templates for nucleic acids or complexes?
No, structure templates are currently supported only for protein monomers.
Written from OpenFold3's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.