BoltzGen Design Space

Design novel protein, peptide, and nanobody binders against any biomolecular target (proteins, DNA, small molecules) using the BoltzGen diffusion model.

BoltzGen unifies structure prediction and binder design in a single all-atom generative model with flexible design constraints.

Developed by Hannes Stärk and team at MIT: description git

Instructions

  1. Prepare a YAML config file specifying your design task. See BoltzGen Docs for examples.
  2. Upload the target structure file (PDB or CIF) corresponding to your design as needed (by the YAML config).
  3. Select the design protocol (chose "protein-anything" for general results).
  4. Specify the number of designs to generate. !Warning: high numbers may take longer to compute than your session allows.
  5. Specify your budget (# of designs wanted in the final diversified set). Note that the budget must be less than or equal to the number of designs.
  6. Click "Run BoltzGen" and wait for the results!
  7. Download your results using the buttons at the bottom:
    • Final Ranked Designs: structures, metrics and report summary of the best designs after filtering and ranking (interaction score and diversity).
    • Intermediate Designs: all generated trajectories before filtering/ranking.
Protocol