HannesStark/boltzgen

What it solves

BoltzGen is a protein design pipeline that generates high-quality binders (proteins, peptides, antibodies, and nanobodies) for specific targets, including other proteins or small molecules. It automates the process of creating new amino acid sequences that can bind to a target structure with high affinity and stability.

How it works

BoltzGen uses a multi-step pipeline to move from a target specification to a ranked set of candidate designs:

1. Design: A diffusion model generates backbone candidates based on a YAML design specification.
2. Inverse Folding: Sequence redesign is performed using models like SolubleMPNN or BoltzIF to determine the best amino acid sequences for those backbones.
3. Folding: The designed binders are refolded with their targets using the Boltz-2 model to verify structural stability.
4. Analysis & Filtering: The pipeline calculates quality metrics and filters the designs to provide a final, diversity-optimized set of ranked candidates.

Who it’s for

It is designed for computational biologists, protein engineers, and researchers working in drug discovery or synthetic biology who need to generate novel binders for specific biological targets.

Highlights

• Versatile Protocols: Dedicated workflows for proteins, cyclic peptides, antibodies, nanobodies, and small-molecule binders.
• Flexible Specifications: Supports detailed YAML configurations for target chains, specific binding sites, secondary structure constraints, and disulfide bonds.
• Symmetry Support: Ability to tie sequence generation for symmetric complexes like homo-dimers.
• Integrated Validation: Combines backbone generation, sequence redesign, and refolding into a single executable pipeline.
• Customizable Filtering: Allows users to quickly rerun filtering steps with different thresholds to refine the final design set.