What is BindCraft?
BindCraft is an open-source, automated pipeline for de novo protein binder design, developed through collaboration between the Correia Lab at EPFL and the Ovchinnikov Lab at MIT. It harnesses the deep-learning capabilities of AlphaFold2 Multimer (AF2M) alongside tools like ProteinMPNN and PyRosetta to design high-affinity protein binders with experimental success rates range from 10% to 100%. BindCraft uses backpropagation through AF2M as one of its core innovations to optimize protein binder design. Backpropagation is the engine that powers BindCraft's ability to turn random sequences into high-affinity binders in very few iterations (Backpropagation-Driven Hallucination). It transforms AlphaFold2 from a predictive tool into a generative one for protein binder design. These binders often achieve nanomolar-range binding affinities without requiring extensive experimental optimization. Unlike rigid-docking approaches, BindCraft accommodates induced-fit conformational changes in the target, enabling precise targeting of dynamic interfaces.
How BindCraft Works?
- Random Sequence Initialization & Co-Folding
- Starts with a random peptide sequence.
- Uses AF2M to co-fold the binder with the target protein, producing an initial (often poor) model.
- AI-Guided Iterative Optimization
- Evaluates designs using multiple scoring metrics:
- Interface interaction strength
- Secondary structure balance (e.g., helix vs. β-sheet ratio)
- Binding site alignment (if specified)
- Applies backpropagation through AF2M to refine sequences toward better scoring models.
- Sequence Refinement & Filtering
- Uses ProteinMPNN to optimize non-interface residues for solubility and foldability.
- Filters out low-quality designs, such as those with structural clashes or poor confidence scores.
- Experimental Validation
- Top computational designs are tested experimentally (e.g., via SPR) and have shown impressive binding affinities against targets like cell-surface receptors, CRISPR-Cas9, and allergens.
Protein Modeling Tools Comparison
| Tool |
Primary Function |
Strengths |
Limitations |
Pharma Applications |
| BindCraft |
De novo protein binder design |
Integrates AF2 Multimer, ProteinMPNN, and PyRosetta
High binder success rates (10–100%)
Works in a one-shot or low-shot setting
Open-source, user-friendly pipeline |
Relatively new, still maturing
Binding success depends on AF2M accuracy |
Design of therapeutic binders
Retargeting viral vectors (e.g., AAVs)
Gene-editing modulation
Allergen or receptor-targeted therapeutics |
| AlphaFold2 |
Protein structure prediction |
State-of-the-art accuracy in 3D structure prediction
Handles single chains well
Openly available |
Limited in predicting protein–ligand interactions
Less accurate with large complexes or dynamics |
Structure-based drug design (SBDD)
Target identification/validation
Protein stability assessment |
| RoseTTAFold |
Structure prediction & protein design |
Faster than AlphaFold2 in some tasks
Integrate into Rosetta suite
Flexible for both monomers & complexes |
Less accurate than AF2 for many cases
Require Rosetta expertise |
Small protein design
Preliminary structure predictions
Educational/research use |
| ProteinMPNN |
Sequence design given a backbone |
Excellent at optimizing non-interface residues
Produce highly foldable, stable sequences
Work well with AF2/BindCraft |
Need an input backbone (cannot generate structures de novo) |
Therapeutic protein engineering
Sequence refinement in binder/antibody design |
| Rosetta |
General protein modeling & design |
Extremely versatile (docking, folding, design)
Industry standard with long track record
Rich plugin ecosystem |
Computationally intensive
Require expert knowledge
Slower than deep learning approaches |
Antibody & enzyme engineering
Protein docking
Vaccine design |
| FoldX |
Protein stability & mutation effect prediction |
Lightweight, fast energy calculations
Good for mutational scanning
User-friendly GUI available |
Less accurate than ML-based tools for full folding
Simplified force fields |
Predicting stability effects of mutations
Rational protein optimization |
Why BindCraft Stands Out?
- High Experimental Success Rates
Average success rate ≈ 46% across diverse and challenging targets, drastically better than conventional methods.
- Fast Progress from Design to Testing
The "one design-one binder" pipeline streamlines computational generation and wet-lab validation.
- Versatility
Works across various therapeutic contexts—from immunomodulation to enhanced delivery systems.
- Accessibility
Its open-source nature and automation make advanced protein design accessible beyond elite computational labs.
Pharmaceutical & Biotech Applications of BindCraft
| Application |
Description |
| Therapeutic Binder Design |
Generates high-affinity binders targeting receptors, allergens, and nucleases for therapeutic interventions (e.g., allergy treatments, gene-editing modulation). |
| Targeted Delivery Systems |
Enables re-targeting of therapeutic vectors, such as AAVs, to specific cell types, improving specificity and reducing off-target effects. |
| Efficient One-Shot Design |
Capable of generating functional binders with very few (often less than 10) candidates, vastly reducing experimental screening burdens. |
| Democratizing Protein Design |
Open-source and user-friendly, BindCraft lowers the barrier for labs to design binders without requiring massive screening facilities. |
Related Services
Binding Protein De Novo Design
Structure Modeling Service
Antibody-Antigen Interaction Modeling Service
Nucleic Acid Binding Protein Modeling Service
Reverse Docking Service
Rigid Docking Service
Peptide Folding Simulation Service
* For Research Use Only.
Related Services
