Aureka open-sources biomolecular foundation model for AI drug discovery

California-based Aureka has released OpenDDE, an open-source, all-atom biomolecular foundation model designed to serve as a structural reasoning core for AI-driven drug discovery. The release makes one of the largest openly available structural reasoning models accessible to academic researchers, startups, and pharmaceutical companies at a time when most competing platforms remain proprietary.

OpenDDE is a 655-million-parameter model — trained at approximately 414,000 GPU-hours — now made freely available to academic laboratories, startups, and large pharmaceutical companies under an Apache-2.0 license. As noted in the release, Aureka's decision to make the platform open source is a bet that open infrastructure will generate broader data diversity and independent validation than closed systems can achieve internally.

OpenDDE uses biomolecular co-folding as its entry point, simultaneously modeling interactions across proteins, nucleic acids, small-molecule ligands, and other biomolecular components. Rather than treating structure prediction as a standalone output, the system is designed as a shared structural reasoning layer connecting sequence, structure, and function — addressing a persistent bottleneck in early-stage discovery: the inability to model complex molecular interfaces, particularly antibody-antigen interactions, with sufficient accuracy to guide downstream design. Aureka describes the architecture as extensible, with future modules planned for de novo molecular design, affinity estimation, and closed-loop experimental feedback.

The company reported competitive antibody-antigen co-folding performance across several public benchmarks, although the results have not yet been independently validated. The company characterizes these results as narrowing the gap with its proprietary IsoDDE model, though independent replication on external benchmarks has not yet been reported.

While companies such as Xaira Therapeutics, Insilico Medicine, Recursion and Isomorphic Labs have largely kept their foundation models proprietary, Aureka is attempting to build an ecosystem around an open platform. Insilico Medicine's Pharma.AI system was recently the focus of a USD 2.75 billion collaboration with Eli Lilly, among a string of other lucrative partnerships. Where those systems retain data and model improvements internally, Aureka is releasing training code, inference pipelines, model checkpoints, and benchmarks — enabling independent validation and community-driven extension. The approach mirrors the dynamics that accelerated large language model development, and represents a calculated wager that openness will compound into durable scientific and commercial advantage.

Aureka's roadmap extends OpenDDE toward a dry-wet closed-loop discovery system, pairing the computational model with a high-throughput automated wet-lab platform integrating single-cell functional screening and automated yeast evolution. The company intends this infrastructure to support complex antibody modalities — including multispecific, internalising, and pH-switch antibodies — with a long-term goal of building differentiated therapeutic pipelines. Whether the open-source release attracts the external contributions needed to sustain that ambition, and whether benchmark performance translates into experimental success, will determine how consequential this foundation proves.

---

This article was generated with AI assistance and reviewed and edited by the AllSci editorial team Explore more at AllSci News: https://allsci.com/news/

---

Spot something wrong? Report an issue with this article