Use BindCraft2 Online
Commercially Available BindCraft2 No-Code Web Server
BindCraft2: Open-Source De Novo Protein Binder Design for Miniproteins, Peptides, VHHs and Antibody Fragments
BindCraft2 (BC2) is the second-generation protein binder design suite from the Pacesa Lab. It is built to make binder design easy and accessible to non-experts without requiring high-throughput screening, and it covers de novo miniproteins, scaffolded binders, cyclic peptides and multistate design in a single workflow.
BC2 combines sequence optimisation with AlphaFold 2, sequence redesign with ProteinMPNN, and evaluation with separate AlphaFold models and structural filters. It returns sequences, predicted complexes, ranked results and measurements of interface, fold and molecular properties. The lab describes it as far more computationally efficient than the original BindCraft, with lower compute cost and runtime, a broader range of targets and binder formats, and an emphasis on experimental success rather than in-silico scores alone.
No publication for BindCraft2 is available yet; the lab lists the preprint as in preparation. This page is based on the Pacesa Lab project page and the BindCraft2 GitHub repository, and it does not report benchmark numbers.
Property | Detail |
|---|---|
Task | De novo protein binder design against protein targets |
Target inputs | Structures in PDB or mmCIF format; disordered regions or linear motifs as FASTA sequences |
Design engine | AlphaFold 2 (via ColabDesign) for optimisation, ProteinMPNN for redesign |
Outputs | Ranked designs, predicted complexes (CIF), CSV tables of metrics and filter outcomes |
Availability | Open source, free for academic and industry use except for hosting services; Google Colab notebook available |
Binder Formats and Design Modalities in BindCraft2
BC2 supports a wide range of binder types, selected with a modality setting:
Miniprotein and large binders: de novo miniproteins, plus large rigid binders of more than 300 residues.
Peptides: linear peptides under 25 residues and cyclic peptides.
Homo-oligomers and multidomain binders: symmetric assemblies and binders with several domains.
Antibody-like formats: VHHs (single-domain antibodies), scFv (two chains, with the linker designed separately) and Fab fragments (heavy and light chains with constant domains).
Ankyrin repeat proteins (ARPs): scaffolded repeat-protein binders.
Conformational modes: induced-fit binders, where the interface changes shape on binding, and fold-switch binders, which adopt different folds when free and when bound. Each uses one target and can combine with a compatible binder format.
Targeting Options and Optimization Objectives
Flexible Target Definition
Structured targets from PDB or CIF files, with optional chain and epitope (hotspot) selection, including chain-prefixed hotspots.
Disordered regions and linear motifs supplied as sequences.
Receptor assemblies using several chains as one target.
Cross-reactive binders across species (for example human and mouse) and single binders optimised against several divergent proteins.
Detargeting to repel chosen off-targets such as close paralogs, and coldspots to keep target regions free.
Molecular glue and bridging designs for induced proximity, and design against one conformational state while detargeting another.
Optional Design Properties
Forced targeting to focus on a named, difficult epitope.
Humanization toward human-like sequence features.
Protease resistance to reduce predicted cleavage, and disulfide staples for rigidity.
Mixed topology to favour beta-sheet content with limited helicity.
Termini control: bring the N and C termini together, or direct them away from the target to keep them free for tags or fusions.
Desperation mode for hard target sites, which keeps iterating settings.
The README notes that these property settings are judged using computational proxies, and objectives can be combined.
BindCraft2 Outputs and Ranking
Each campaign runs until the requested number of accepted designs is reached, with no limit on design attempts. Results are organised into three folders: trajectories (attempts and optimisation records), refolded designs (redesigned sequences, complexes and filter outcomes) and ranked designs (accepted designs). Accepted designs are ranked by i_pDAE, a distance-masked interface confidence score, and you can re-rank by another measurement or re-filter with new thresholds without rerunning the design. Each campaign also records its resolved settings, model choices and checkpoint hashes.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Binder Discovery with BindCraft2 on Tamarind Bio
Therapeutic binder design: generate miniproteins, VHHs, scFv or Fab fragments against a target of interest.
Peptide design: design linear or cyclic peptide binders, including against disordered regions given as sequences.
Selectivity and cross-reactivity: design one binder for several species or proteins, or repel close paralogs with detargeting.
Hard epitopes: focus designs on a chosen site with forced targeting and hotspots.
No local setup: run design campaigns without installing the Linux toolchain or managing GPUs yourself.
How to Use BindCraft2 on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and select BindCraft2.
Provide your target: upload a PDB or CIF structure, or enter a sequence for a disordered target, and select the chain(s).
Choose a modality: pick a binder format such as miniprotein, peptide, cyclic peptide, VHH, scFv or Fab, and set binder length where applicable.
Set hotspots and objectives: optionally specify epitope hotspots, off-targets to detarget, and properties such as humanization or protease resistance.
Choose the number of designs: set how many accepted designs you want and submit the job.
Review results: download the ranked designs, predicted complexes and metric tables, and shortlist candidates by interface confidence.
Parameters shown on Tamarind may differ from the full command-line options, so refer to the tool page for what is exposed.
Limitations and Experimental Validation
BindCraft2 handles protein-protein interactions only; ligands, post-translational modifications, small molecules and nucleic acids are not accounted for.
Only natural amino acids are supported in the binder and the target.
Ranking reflects binding probability, not predicted affinity, and confidence scores are not binding affinities.
Some target sites are difficult and may yield few passing designs.
Target and binder size are limited by GPU memory.
Outputs are computational designs: binding, selectivity and biological behaviour require experimental validation, and the lab expects roughly 2 to 20 designs to be screened in the wet lab.
Source: Pacesa Lab, BindCraft2 project page (pacesalab.com/bindcraft) and GitHub repository (github.com/PacesaLab/BindCraft2). The BindCraft2 preprint is listed as coming soon. The original BindCraft is described in Pacesa, Nickel, Schellhaas et al., "One-shot design of functional protein binders with BindCraft," Nature (2025).