AI Models for Protein Structure Prediction
OpenDDE
OpenDDE is an open-source, all-atom biomolecular foundation model developed by Aureka AI Research. Rather than predicting a protein structure in isolation, OpenDDE uses co-folding to model complete biomolecular assemblies, including proteins, nucleic acids, small-molecule ligands, ions, and covalent interactions, within a single structural framework. It is designed for antibody-antigen structure prediction, protein-ligand modeling, molecular design, and structure-conditioned optimization. Choose OpenDDE when you want to understand or design a biomolecular complex, particularly in a drug-discovery context, rather than simply predict the structure of a single protein.
- All-atom co-folding: Models complete biomolecular assemblies within a unified structural framework.
- Multiple molecular components: Supports proteins, nucleic acids, small-molecule ligands, ions, and covalent interactions.
- Complex-focused prediction: Particularly suited to studying interactions such as antibody-antigen and protein-ligand complexes.
- Structure-guided molecular design: Supports applications in molecular design and structure-conditioned optimization.
- Open-source foundation: Provides an open framework for researchers to run, evaluate, and build upon the technology.
Created by
Aureka AI Research
Research & Benchmarking
OpenDDE reference publication
FoldBench benchmark
RESEARCH ARTICLE • PREPRINT
Folding, Reasoning, and Scaling with Open-source Drug Discovery Engine
arXiv • 2026 • doi: 10.48550/arXiv.2607.03787
Introduces OpenDDE, an open-source, all-atom biomolecular foundation model that uses co-folding as the entry point to an AI-driven drug discovery engine. The work presents OpenDDE as a shared structural reasoning framework for biomolecular complex prediction and future applications in molecular design and optimization.
RESEARCH ARTICLE • OPEN ACCESS
Benchmarking all-atom biomolecular structure prediction with FoldBench
Nature Communications • 2026 • doi: 10.1038/s41467-025-67127-3
Introduces FoldBench, a low-homology benchmark of 1,522 biological assemblies spanning nine prediction tasks. The benchmark includes 172 antibody-antigen interfaces and identifies antibody-antigen prediction as a particularly challenging task for current all-atom structure prediction methods.
DNASTAR Benchmarks
The limiting hardware factor for OpenDDE predictions is GPU memory (VRAM).
Cloud vs. Local Performance
DNASTAR benchmark comparing OpenDDE prediction times on a local computer with an NVIDIA RTX 6000 (96 GB VRAM) and Amazon cloud with an NVIDIA Tensor Core GPU (48 GB VRAM).
The benchmark graph reports measured prediction times for the listed protein structures. Orange X markers indicate local-computer predictions that failed after measured runtime; green X markers indicate Amazon-cloud predictions that failed after measured runtime.
