Leading protein design researchers have released BoltzDesign1, an open-source tool that extends AI-powered protein-binder design to targets such as small molecules, nucleic acids, metals and PTMs.

The work (preprint here) is led by breakout star Yehlin Cho, a graduate student in Sergey Ovchinnikov's lab at MIT. Sergey is a leading researcher in the protein hallucination approach with ColabDesign, which uses backpropagation through structure prediction models to design new proteins. The collaboration includes Martin Pacesa and Bruno Correia from EPFL—the team behind BindCraft, which automated and improved upon the ColabDesign approach.

A Significant Technical Advance

While BoltzDesign1 builds on these earlier tools, it represents a major leap forward in capability. Instead of using AlphaFold2 for protein hallucination, it leverages Boltz-1, an open-source implementation of AlphaFold3's all-atom structure prediction capabilities. This enables the design of proteins that bind not just to other proteins, but to small molecules, nucleic acids, metal ions, and even post-translationally modified residues.

The preprint reports encouraging in silico results:

• High success rates (up to ~90% for the four benchmark ligands under relaxed criteria) for small molecule binder design, comparing favorably to existing methods
• Greater structural diversity in generated designs
• Successful design of binders for DNA, RNA, metal ions, and proteins with covalent modifications
• The ability to model flexible ligand conformations during the design process

Remarkably, the team achieved these results while dramatically reducing computational costs. By cleverly optimizing the protein design process at an earlier stage in the AI model's calculations, they avoid the expensive computations that other methods require—making BoltzDesign1 both powerful and efficient.

Early Days, Promising Future

BoltzDesign1 is still in active development. The pipeline continues to be refined, and the computationally designed binders await experimental validation. However, there's strong reason for optimism: both ColabDesign and BindCraft have proven highly successful in practice, and BoltzDesign1's impressive in silico performance—which has historically been a good predictor of experimental success—suggests this tool will follow suit.

The MIT-licensed code is freely available, and the team has provided a Google Colab notebook for researchers who want to explore the tool's capabilities.

At Ariax, we're genuinely excited about BoltzDesign1's potential. Our philosophy has always been clear: we offer only the most powerful, open-source tools that have been experimentally validated and can be used by any researcher, regardless of their computational background. While BoltzDesign1 awaits that final experimental validation, we fully expect it will join our platform once those results arrive. The combination of this team's track record, the tool's technical innovations, and its outstanding computational performance all point toward another breakthrough in democratizing protein design. Until then, we encourage adventurous researchers to explore this promising new addition to the open-source protein design ecosystem.