More Possibilities for Protein Design
BindCraft changed what a small research team could do. A researcher could describe a target, run an automated design campaign, and take a manageable shortlist of protein binders to the laboratory. We called it a DeepSeek moment in AI drug discovery, and it inspired us to build Ariax Bio to make tools like it accessible without requiring scientists to become GPU infrastructure specialists.
BindCraft2 expands what researchers can design. Developed by Martin Pacesa and the BC2 team, it brings a much faster campaign engine, a wider range of therapeutic formats, and more ways to design around the biology of an experiment. It builds on the approach that made the original BindCraft work, giving researchers more options for choosing a binder format, shaping its interactions, and testing new ideas.
A Foundation That Has Earned Its Place
Nearly two years after its first release, BindCraft remains our benchmark for practical miniprotein design. Its appeal has endured because the evidence goes beyond attractive predicted structures.
The original Nature study reported experimental success rates of 10–100% across its tested targets, including functional binders for receptors, allergens, and other challenging proteins. Those are results from that study, not a promised hit rate for a new campaign. Independent results also mattered: in Adaptyv's second EGFR design competition, six of the seven successful de novo binders were designed with BindCraft.
Strong experimental results and an automated workflow made BindCraft a practical starting point for researchers designing binders against new targets.
BindCraft2 retains the core strategy: optimize a binder through AlphaFold2, redesign sequences with ProteinMPNN, then repredict and filter the resulting candidates. The design, redesign, and validation steps now sit within a broader system of modality-specific presets and biological objectives. Familiar logic has been improved and extended into a much more capable tool.
A Major Step Forward in Speed
Protein design is a search. Faster execution means more opportunities to explore a binding site, compare formats, and revise an experiment while the biological question is still fresh.
BindCraft2 tackles the overhead throughout that search. It reuses compiled calculations, groups compatible sequence lengths to avoid unnecessary recompilation, prepares upcoming calculations in advance, and can run multiple workers on a GPU when memory allows. Its campaign machinery also coordinates parallel work and resumes interrupted searches.
These changes make speed a central part of the upgrade, alongside the new scientific capabilities. They reduce time spent preparing the same calculations and help more of the available compute go into designing candidates. Runtime still depends on target size, binder format, filters, hardware, and how readily a campaign finds accepted designs.
On Ariax, the same workflow is available without installing the engine or maintaining its GPU environment. Choose your compute preferences, start a campaign, and return to the project to review progress and results. Turbo Mode offers additional parallel compute when you want to shorten the wait.
More Ways to Build a Binder
The original BindCraft made de novo miniprotein design broadly usable. BindCraft2 takes that design approach into a much wider set of formats:
- Fab: antibody fragments with heavy- and light-chain variable and constant domains.
- scFv variable domains: paired antibody variable domains. BC2 models these as two chains; a connecting linker must be designed separately.
- VHH: single-domain antibody binders based on a supplied scaffold.
- Large binders and miniproteins: de novo proteins across a broader range of sizes.
- Ankyrin Repeat Protein (ARP): binders built on an ankyrin-repeat scaffold.
- Linear and cyclic peptides: short binders, including designs intended for head-to-tail cyclization.
Specialized setups also support homo-oligomers and multidomain binders. These are available within BindCraft2's setup workflow when the experiment calls for them.
The distinction matters scientifically. Choosing a format defines the starting structure, editable regions, optimization settings, and acceptance criteria. A short peptide that folds on binding should not be judged in exactly the same way as an independently folded miniprotein or a scaffolded antibody fragment.
Design Around the Experiment
A binder's format is only the beginning. Often the harder question is what it needs to recognize, avoid, or leave accessible.
BindCraft2 can optimize against multiple targets, discourage binding to specified off-targets, focus contacts on a chosen epitope, and avoid selected target regions. Compatible optional properties can favor human-like sequence features, protease stability, disulfide formation, or useful arrangements of the termini. Conformational objectives explore changes between free and bound states.
That opens up more useful experiments: a binder intended to recognize related targets, a candidate designed away from a competing interface, or a protein whose attachment points need to remain exposed for a fusion construct.
These objectives use computational measurements. They help define and select candidates; binding, selectivity, stability, and therapeutic behavior still need experimental testing. The Ariax setup page explains each option and checks supported combinations before launch.
The Same Ariax Workflow
You do not need to learn a new platform to use the new engine.
Upload a target structure or find one in the Protein Data Bank, select the relevant chains, and choose a format. Start with its defaults, then open optional settings only when your experiment needs them. Set the number of accepted designs you want and, for a bounded pilot, a maximum number of design attempts.
The project page keeps settings, status, compute usage, structures, and downloads together. Review accepted designs in Final Results, inspect rejected candidates and earlier trajectories when you need to understand the search, and download the CSV tables, individual structures, or results archive. Pause a campaign and restart from its saved progress, or use Clone & Reuse to set up a new experiment.
The BindCraft2 project guide walks through the options and explains how to interpret the results.
Free to Use on Ariax. Pay Only for Compute.
Academic and commercial researchers can use BindCraft2 on Ariax Bio with no software fee, subscription, or separate commercial license to obtain. You pay only for the compute your campaigns use.
BindCraft2's upstream license is source-available and restricts third-party hosting. Ariax handles the licensing needed to offer this hosted workflow.