Aureka, an innovative TechBio corporation specializing in advanced infrastructure for AI-powered biologics discovery, officially announced the debut of its Open Drug Discovery Engine (OpenDDE). The pioneering, open-source, all-atom biomolecular foundation model is specifically built to operate as the core structural reasoning engine for next-generation drug discovery workflows.
Rather than isolating structure prediction as a standalone metric, OpenDDE leverages biomolecular co-folding to model complex interactions across nucleic acids, proteins, small-molecule ligands, and additional biomolecular variables. This unified setup creates a collaborative structural reasoning layer tailored for sequence-structure-function modeling. While delivering robust complex structure prediction capabilities today, the engine establishes a framework for future applications in affinity estimation, de novo molecular design, structure-conditioned optimization, and continuous experimental loops.
Also Read: Sino Biological Unveils XPressMAX™ Cell-Free Protein Synthesis Kit to Accelerate AI-Driven, High-Throughput Antibody Drug Discovery
Across various in silico testing environments, OpenDDE delivers competitive co-folding capabilities, effectively closing the gap with highly regarded, proprietary IsoDDE-level standards. Concurrently, it offers the broader scientific and biomedical research community a fully transparent and reproducible framework.
“OpenDDE begins with open, all-atom co-folding and structural reasoning. On selected in silico benchmarks, it shows competitive performance that narrows the gap with reported IsoDDE-level results. We view this release as an early foundation toward a broader drug discovery engine: a system that can progressively connect structure prediction, molecular design, affinity estimation, and experimental feedback to support more informed exploration of disease- and target-relevant molecular space.” – Will Hua from Aureka AI Research.
Key Technological Innovations of OpenDDE
The engine incorporates three core pillars that elevate its capabilities within the TechBio landscape:
- Atomic Latent Reasoning Over Biomolecular Tokens: OpenDDE pioneers latent reasoning methodologies within biomolecular modeling frameworks. The system systematically optimizes representations of chemical context, local geometry, and cross-molecular interfaces prior to commencing all-atom structure generation.
- A Folding-Centered, Highly Extensible Architecture: Although the initial launch emphasizes complex structure prediction, the shared infrastructure is pre-engineered to effortlessly assimilate structure-conditioned modules, affinity predictions, and de novo molecular configurations.
- Established Scaling Laws and Data Distillation: Aureka evaluated structural scaling across training, data, inference, and model-parameter axes. This research presents verifiable, practical pathways for the iterative improvement of foundational biomolecular frameworks.
Validating Benchmark Excellence in Antibody-Antigen Co-Folding
Aureka’s comprehensive technical documentation indicates that OpenDDE yields outstanding antibody-antigen co-folding results across three standardized evaluations. When measuring top-ranked selections, OpenDDE secured a 51.0% success rate on PXMeter-AB, 70.0% on FoldBench-AB, and 66.4% on the newly curated 2026ARK-AB benchmark.
Under oracle selection criteria, these success metrics increased sharply to 65.9%, 81.9%, and 80.1%, respectively. This data reflects an exceptionally robust latent sampling volume and confirms that additional improvements are achievable via refined candidate ranking and confidence calibration.
These milestones hold profound implications for therapeutic pipelines, given that antibody-antigen interfaces are traditionally flexible, highly complex, and chemically varied. OpenDDE elevates low-threshold recovery metrics while significantly strengthening medium- and high-quality DockQ parameters, illustrating superior capacity in modeling strict binding geometry rather than simply rendering minimally viable complexes.
Propelling Biomolecular AI into the Scaling Era
OpenDDE has been constructed using 655 million trainable parameters and has needed a huge amount of 414,000 GPU hours of computational effort, equaling the computation that would take around 54 years on a single computer. This immense scale highlights a key change in the contemporary life sciences industry: AI applied in biology is not limited to algorithmic challenges anymore, but it is an infrastructural challenge requiring colossal computational power, automation, and engineering skills.
Aureka’s initial analytics validate explicit scaling trajectories, showing that expanded training corpora, larger neural frameworks, advanced inference sampling, and post-training optimization translate reliably into better biological reasoning and structure prediction. Consequently, biomolecular AI is entering a historical scaling paradigm reminiscent of the architectural shift that revolutionized large language models (LLMs).
Empowering Global Science via Open-Source Access
In an attempt to make biomolecular modeling accessible to all educational institutions globally, startups, individual research laboratories, and multinational companies, Aureka has come up with OpenDDE under the permissive Apache-2.0 license. The model includes comprehensive training and inference code, checkpoints, and benchmarking, which can be used for validation and collaboration.
The open-source components can be accessed via:
- GitHub: https://github.com/aurekaresearch/OpenDDE
- Hugging Face: https://huggingface.co/aurekaresearch/OpenDDE
- Official Site: https://aurekaresearch.github.io/OpenDDE-Website/
Constructing the Future of TechBio Infrastructure
OpenDDE stands as a crucial cornerstone for Aureka’s overarching mission. By combining autonomous antibody-design agents with high-throughput single-cell functional screenings and automated yeast evolution, the company is generating a continuous, high-content experimental data flywheel for functional antibody discovery.
Moving forward, Aureka intends to scale this architecture to support conformational ensemble modeling, de novo design, and holistic scientific world modeling. This expanding infrastructure is tailored to fuel the next wave of antibody discovery across highly intricate modalities including multi-specific, epitope-specific, internalizing, and pH-switch antibodies advancing toward the ultimate objective of engineering highly differentiated, First-in-Class therapeutics.
