Use OpenDDE Online

OpenDDE: Open-Source All-Atom Co-Folding for Antibody-Antigen and Protein Complex Structure Prediction

OpenDDE (Open Drug Discovery Engine) is an open-source, all-atom biomolecular foundation model that uses co-folding to predict the 3D structures of protein, nucleic-acid and small-molecule complexes. It was developed by the OpenDDE Project at Aureka AI Research, and its code, inference pipelines, checkpoints and benchmarks are released under the Apache-2.0 license.

Characterizing how proteins, nucleic acids, ligands and other cellular components interact is fundamental to understanding disease and to designing therapeutics. AlphaFold2 transformed monomer prediction, and AlphaFold3 extended the idea to joint atomic-level modeling of complexes. The authors note, however, that incremental gains in co-folding accuracy have begun to slow, and that strong closed systems limit independent validation and community extension.

OpenDDE responds with an open model that the authors describe as reaching accuracy comparable to the closed IsoDDE system on several co-folding benchmarks, with the largest improvements on antibody-antigen generalization.

How OpenDDE Predicts Biomolecular Complexes with Atomic Latent Reasoning

OpenDDE is a coarse-to-fine model that unifies structure prediction and conditional design in one diffusion framework. The paper highlights three contributions:

1. Atomic latent reasoning: a Pairformer-style trunk builds residue-level representations from sequence, atom, MSA, template and constraint features. These are expanded into chemically meaningful structural tokens (backbone, side chain, nucleic-acid base, ligand atoms), and a Structural Refiner reasons over them before coordinates are generated.

2. Diffusion-based coordinate generation: a denoising network predicts all-atom positions conditioned on the refined states, with confidence and distance heads providing quality estimates.

3. Scaling laws and data distillation: the authors study scaling along model-size and data axes and revise distillation strategies for monomeric and multimeric structures to improve robustness and training efficiency.

OpenDDE Benchmark Results for Antibody-Antigen Docking Accuracy

Antibody-antigen interfaces are a stringent test because they are flexible and chemically diverse. The paper reports DockQ success rates (acceptable quality at DockQ above 0.23) on three benchmark collections, using ranking-based and oracle selection of the best sample.

Newly released structures

The authors curated 2026ARK-AB, 164 recent PDB complexes covering 159 unique antibody-antigen interface clusters, to test generalization to low-homology systems. On it OpenDDE scores 66.4%, compared with ESMFold2 (51.0%), Protenix-v1 (50.0%), Boltz-1 (25.0%), OpenFold3 (21.9%) and Chai-1 (16.7%).

Broader FoldBench results

The paper reports a protein monomer LDDT of 0.890, a protein-protein score of 0.769 and an antigen-antibody score of 0.700 (versus 0.581 for ESMFold2, 0.488 for AlphaFold3 and 0.485 for Protenix-v1). It also reports competitive scores for RNA monomers (0.660), protein-ligand complexes (0.601) and protein-RNA complexes (0.735).

Test-time scaling

Oracle rates well above ranked rates suggest the model often generates a high-quality conformation that is not ranked first, so better confidence estimation and additional sampling seeds can improve results.

What is Tamarind Bio?

Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.

The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:

• A user-friendly graphical interface for setting up and launching experiments

• A robust API for integration into existing research pipelines

• An automated system for managing and scaling computational resources

Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.

Accelerating Discovery with OpenDDE on Tamarind Bio

• Antibody-antigen modeling: predict binding interfaces for therapeutic antibodies, a regime where the paper reports its largest gains.

• Protein-protein complexes: model assemblies and interfaces as a structural basis for target discovery and validation.

• Epitope modeling and ensemble sampling: the authors position co-folding as a reusable layer for epitope prediction and conformational sampling.

• Foundation for design: the authors describe the architecture as a base for future de novo design, hit-to-lead optimization and affinity estimation.

How to Use OpenDDE on Tamarind Bio

1. Open the tool: log in to tamarind.bio and open OpenDDE.

2. Provide the complex: enter protein, DNA or RNA sequences, and ligands or ions, as the tool accepts them.

3. Choose a checkpoint: if exposed, select the general or antibody-antigen model.

4. Set sampling options: if exposed, increase the number of seeds or samples, since the paper shows that more samples raise the chance of a high-quality prediction.

5. Run the prediction: submit the job.

6. Download outputs: retrieve the predicted structures and confidence scores.

7. Interpret results: inspect interface geometry and confidence, and compare several candidates rather than only the top-ranked one.

8. Use downstream: take the models into design, analysis or experimental planning.

Things to Keep in Mind

• The paper focuses mainly on protein-protein and antibody-antigen interactions; the authors do not claim optimization for ligand docking, virtual screening or affinity ranking.

• Comparisons with IsoDDE are limited by the little public information available about that system.

• Top-ranked predictions are not always the best sample, so confidence estimation remains an area for improvement.

• Design, affinity prediction and ensemble modeling are described as future extensions, not validated components.

Source: OpenDDE Project, Aureka AI Research, "Folding, Reasoning, and Scaling with Open-source Drug Discovery Engine" (technical report). Code: https://github.com/aurekaresearch/OpenDDE. Model: https://huggingface.co/aurekaresearch/OpenDDE.