Use BoltzGen Online
Commercially Available BoltzGen No-Code Web Server
BoltzGen: All-Atom Generative Model for Protein and Peptide Binder Design
BoltzGen is an all-atom generative model for designing proteins and peptides that bind diverse biomolecular targets. It was developed by Hannes Stark, Felix Faltings, MinGyu Choi, Yuxin Xie and colleagues from MIT, Boltz, CTU Prague, IOCB Prague, NVIDIA, UC Irvine, UCSF and other institutions, in the paper "BoltzGen: Toward Universal Binder Design". Code, model weights, training code and designs are released under the MIT license.
Designing binders is difficult because a good design must fold, fit a target surface and satisfy chemical constraints at once. Many methods handle only one binder type or one kind of target, and separate tools are often needed for design and for structure prediction.
BoltzGen unifies design with structure prediction in a single model and reaches state-of-the-art folding performance according to the authors, matching Boltz-2 on their test set. Generation can be steered with a specification language covering covalent bonds, structure constraints and binding sites.
Property | Detail |
|---|---|
Model type | Single all-atom diffusion model built on the Boltz-2 architecture |
Design targets | Proteins, peptides, nanobodies, disordered proteins and small molecules |
Training data | PDB, AlphaFold DB and Boltz-1 distillation |
Training tasks | Folding, binder design, motif scaffolding, unconditional design |
Wet-lab validation | Eight campaigns covering 26 targets |
License | MIT |
Why Universal Binder Design Matters
One model that handles many binder formats and target types reduces the need to stitch together separate pipelines. The authors validated designs across nanobodies, miniproteins, linear and cyclic peptides, and targets ranging from disordered proteins to small molecules.
How BoltzGen Designs Binders
Geometric residue encoding: Designed residue identity is encoded with 14-atom placeholder representations rather than discrete labels.
Generation: The diffusion model generates designs under user-specified constraints.
Optional inverse folding: BoltzIF, a SolubleMPNN-like model, can redesign sequences.
Refolding and metrics: Designs are refolded with Boltz-2 and scored.
Selection: A quality-diversity algorithm ranks designs and selects a diverse set.
BoltzGen Experimental Results
Testing used multiple laboratories with SPR, BLI, fluorescence, CD, SEC, antimicrobial, hemolysis and in-cell assays.
Campaign | Reported outcome |
|---|---|
Novel targets (10, under 30% identity to bound PDB proteins) | Nanobody hits for 6 of 10 targets; miniprotein hits for 5 of 10 (15 top designs per target per modality) |
Rag GTPase peptides | 8 of 27 linear and 23 of 26 cyclic peptides showed concentration-dependent SPR binding |
GyrA antimicrobial peptides | 352 of 1,808 designs (19.5%) inhibited E. coli growth |
Heme binder d1 | Kd of 697 plus or minus 663 nM |
Melittin binders mel2 and mel3 | Kd of 0.41 and 4.4 micromolar |
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 Discovery with BoltzGen on Tamarind Bio
Nanobody and miniprotein binders: Design binders against targets with limited prior structural precedent.
Peptide design: Generate linear or cyclic peptides, as in the Rag GTPase and bioactive peptide campaigns.
Small-molecule binding proteins: Design proteins for ligands such as heme, as demonstrated by the authors.
Disordered targets: Create binders for disordered proteins such as NPM1 and NUP98.
How to Use BoltzGen on Tamarind Bio
Log in and open the tool: Sign in at tamarind.bio and open BoltzGen.
Define the target: Provide the target structure or sequence and, if exposed, the binding site or constraints using the design specification.
Choose the binder type: If exposed, select the modality, such as protein, peptide or nanobody, and the design length.
Set the number of designs: If exposed, choose how many designs to generate and how many to keep after filtering.
Run the job: Submit and wait for generation, refolding and ranking.
Download outputs: Retrieve ranked designs and their metrics.
Interpret and use downstream: Inspect predicted complexes, then order top designs for binding assays.
Things to Keep in Mind
The authors caution against treating BoltzGen as zero-shot or plug-and-play; inspect outputs and iterate.
Therapeutic success also depends on selectivity, developability and target-specific properties that the model does not fully address.
For binders of length 73 to 76, generation can collapse to ubiquitin for some targets.
Some campaigns gave weak affinities, for example rucaparib binders in the 43 to 152 micromolar range.
Source: Stark H, Faltings F, Choi M, Xie Y, et al. BoltzGen: Toward Universal Binder Design. Paper PDF. Code: github.com/HannesStark/boltzgen.