Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
[Preprint]. 2025 May 6:2024.11.19.624167.
doi: 10.1101/2024.11.19.624167.

Boltz-1 Democratizing Biomolecular Interaction Modeling

Jeremy Wohlwend  1   2 , Gabriele Corso  1   2 , Saro Passaro  1   2 , Noah Getz  1   2 , Mateo Reveiz  1   2 , Ken Leidal  3 , Wojtek Swiderski  3 , Liam Atkinson  4 , Tally Portnoi  1   2 , Itamar Chinn  1   2 , Jacob Silterra  1   2 , Tommi Jaakkola  1   2 , Regina Barzilay  1   2
Affiliations

Boltz-1 Democratizing Biomolecular Interaction Modeling

Jeremy Wohlwend et al. bioRxiv. .

Abstract

Understanding biomolecular interactions is fundamental to advancing fields like drug discovery and protein design. In this paper, we introduce Boltz-1, an open-source deep learning model incorporating innovations in model architecture, speed optimization, and data processing achieving Alphafold3-level accuracy in predicting the 3D structures of biomolecular complexes. Boltz-1 demonstrates a performance on-par with state-of-the-art commercial models on a range of diverse benchmarks, setting a new benchmark for commercially accessible tools in structural biology. Further, we push the boundary of capabilities of these models with Boltz-steering, a new inference time steering technique that is able to fix hallucinations and non-physical predictions from the models. By releasing the training and inference code, model weights, datasets, and benchmarks under the MIT open license, we aim to foster global collaboration, accelerate discoveries, and provide a robust platform for advancing biomolecular modeling.

PubMed Disclaimer

Figures

Figure 1:
Figure 1:
Example predictions of Boltz-1 on targets from the test set.
Figure 2:
Figure 2:
2D representation of the difference between AlphaFold3 reverse diffusion and Boltz-1 reverse diffusion with our Kapsch interpolation. Colors indicate correspondence between different points. Even though the prediction of the denoising model is “perfect” according to the aligned MSE loss, the unaligned interpolation may lead to poor structures fed to the next reverse diffusion step.
Figure 3:
Figure 3:
Diagram of the architecture of Boltz-1. The critical difference with AlphaFold3 lies in the confidence model, which now not only has a PairFormerModule but follows a full trunk composition and is fed features coming from the denoising model through the recursive updates.
Figure 4:
Figure 4:
Forwards runtime of trifast kernel compared to compiled PyTorch and DeepSpeed kernel.
Figure 5:
Figure 5:
Visual summary of the performance of AlphaFold3, Chai-1, Boltz-1 and Boltz-1x on the test set.
Figure 6:
Figure 6:
Visual summary of the performance of AlphaFold3, Chai-1, Boltz-1 and Boltz-1x on the CASP15 benchmark.
Figure 7:
Figure 7:
Examples of some failure modes of Boltz-1 leading to unphysical poses, on the left, and the fixed poses resulting from Boltz-1x, on the right.
See this image and copyright information in PMC

References

    1. Abramson Josh, Adler Jonas, Dunger Jack, Evans Richard, Green Tim, Pritzel Alexander, Ronneberger Olaf, Willmore Lindsay, Ballard Andrew J, Bambrick Joshua, et al. Accurate structure prediction of biomolecular interactions with alphafold 3. Nature, pages 1–3, 2024. - PMC - PubMed
    1. Ahdritz Gustaf, Bouatta Nazim, Floristean Christina, Kadyan Sachin, Xia Qinghui, Gerecke William, O’Donnell Timothy J, Berenberg Daniel, Fisk Ian, Zanichelli Niccolò, et al. Openfold: Retraining alphafold2 yields new insights into its learning mechanisms and capacity for generalization. Nature Methods, pages 1–11, 2024. - PMC - PubMed
    1. Bansal Arpit, Chu Hong-Min, Schwarzschild Avi, Sengupta Soumyadip, Goldblum Micah, Geiping Jonas, and Goldstein Tom. Universal guidance for diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 843–852, 2023.
    1. Berman Helen M, Westbrook John, Feng Zukang, Gilliland Gary, Bhat Talapady N, Weissig Helge, Shindyalov Ilya N, and Bourne Philip E. The protein data bank. Nucleic acids research, 28(1):235–242, 2000. - PMC - PubMed
    1. Biasini Marco, Schmidt Tobias, Bienert Stefan, Mariani Valerio, Studer Gabriel, Haas Jürgen, Johner Niklaus, Schenk Andreas Daniel, Philippsen Ansgar, and Schwede Torsten. Openstructure: an integrated software framework for computational structural biology. Acta Crystallographica Section D: Biological Crystallography, 69(5):701–709, 2013. - PMC - PubMed

Publication types

LinkOut - more resources