OpenFold Consortium Releases OpenFold3-Preview, Aiming for AlphaFold3 Parity with Open-Source Model

The OpenFold Consortium has announced the preview release of OpenFold3 (OF3p), an open-source deep learning model designed for biomolecular structure prediction. This new model aims to achieve "full AF3-parity for every modality," directly challenging the capabilities of DeepMind's AlphaFold3, as stated by Mohammed AlQuraishi, a key developer. The release is intended to foster community feedback and enable development within the OpenFold ecosystem.

OpenFold3-preview accurately predicts the three-dimensional structures of complex proteins and their interacting molecules, including nucleic acids and small molecule ligands. Trained on over 300,000 publicly available experimental structures and an additional 13 million synthetic structures curated by OpenFold, the model significantly accelerates in silico screening for drug discovery and materials science R&D. Its development has involved an investment of approximately $17 million.

A critical distinction of OpenFold3 is its Apache 2.0 license, which allows for broad academic and commercial use. This contrasts with AlphaFold3, which is available for limited academic use but not licensed for industrial applications. Woody Sherman, Chief Innovation Officer at Psivant Therapeutics and OpenFold Consortium Executive Committee Chairperson, highlighted this as a "big step forward in terms of the democratization of AI structural-biology tools."

The model's development was led by the AlQuraishi Lab at Columbia University, the Bioresilience Program at Lawrence Livermore National Laboratory, and the Steinegger Lab at Seoul National University. The consortium is actively working towards achieving full parity with AlphaFold3's functionality, with a full release planned in the coming months. Developers are encouraged to begin integrating the preview into their workflows.

Major commercial enterprises are already committing to leveraging OpenFold3 to accelerate their R&D efforts. Novo Nordisk plans to adapt the model for internal pipelines, while Outpace Bio will use it to generate new cell therapies. Bayer Crop Science intends to apply OpenFold3 to study proteins for crop protection, and Cyrus Biotechnology will use it for designing enzyme-based drugs.

Built with PyTorch and available via NVIDIA NIM, OpenFold3 is designed for adaptability and extensibility. Mohammed AlQuraishi noted that its modularity allows biopharma and materials science companies to modify the model to interpret their native data formats, lowering the barrier to entry for widespread adoption.