Iterative protein binder design loop with optimization cycles converging on a target structure

Iterative binder design

BindCraft iterative protein binder design

Iterative binder design with tightly integrated scaffold generation and sequence scoring for higher-quality candidates

End-to-end

Scaffold + sequence + validation in one loop

500-2K

Designs per run

Iterative

Scaffold + sequence co-optimization

What BindCraft is

Integrated scaffold and sequence design in one loop

BindCraft combines backbone scaffold generation with iterative sequence optimization in a single pipeline. Rather than generating a backbone first and designing sequences separately (as in the RFdiffusion + ProteinMPNN workflow), BindCraft co-optimizes structure and sequence together, directly scoring for predicted target engagement at each iteration.

This integrated approach produces candidates with stronger initial interface metrics and higher predicted binding confidence. The tradeoff is throughput: BindCraft generates hundreds to low thousands of candidates per run, compared to RFdiffusion's tens of thousands of backbones.

Design philosophy

Quality over quantity

RFdiffusion prioritizes throughput: cast a wide net and filter. BindCraft prioritizes per-candidate quality: generate fewer designs, but each one has already been iteratively refined.

In practice, BindCraft candidates pass downstream structural validation at a higher rate, meaning a smaller initial pool can yield a comparable number of synthesis-ready sequences.

Iterative process

How BindCraft designs binders

1 Initialize scaffold geometry from target binding site

2 Generate initial backbone and sequence jointly

3 Score predicted interface quality and binding geometry

4 Iteratively refine scaffold and sequence toward scoring objectives

5 Output optimized binder with integrated confidence metrics

In combination

How BindCraft complements RFdiffusion and Boltzgen

Ranomics runs BindCraft in parallel with RFdiffusion and Boltzgen. Each approach explores different regions of design space. RFdiffusion generates diverse backbone topologies at high throughput. Boltzgen samples alternative binding geometries via flow-based generation. BindCraft produces fewer but more refined candidates with stronger predicted interfaces.

Running all three generators in parallel maximizes the diversity and quality of the candidate pool entering experimental screening. For targets where a specific binding mode is required (e.g., blocking a receptor interface), BindCraft's ability to impose direct geometric constraints on the output scaffold is particularly valuable.

Comparison

Three generators, one filtering pipeline

Dimension	RFdiffusion	BindCraft	Boltzgen
Throughput	10,000-50,000 designs/run	500-2,000 designs/run	Hundreds to thousands/run
Approach	Sequential: backbone then sequence	Iterative: co-optimized	Flow-based generative sampling
Strength	Broad exploration, diverse topologies	Higher per-candidate quality, specific binding modes	Alternative geometries beyond diffusion training distribution
Sequence design	Separate (ProteinMPNN)	Integrated in loop	Separate (ProteinMPNN)

Run a BindCraft campaign for your target

We run BindCraft and RFdiffusion in parallel to maximize your candidate pool. Tell us about your target and we will recommend the right campaign structure.

Start a project →