How to design a dsRNA binder protein using Boltzgen?

[bee8617]

Hello,

I am planning to design binders (100-250 residues) that will bind to dsRNA without any sequence-specificity. The binders must distinguish between dsDNA, ssRNA and dsRNA. Is Boltzgen good to start with? I did not find any specialized module to refer to for designing dsRNA or RNA-binders. How should the .yaml file be different from the ones used for DNA-binders ? What metrics should I use for scoring and what computational pipeline should I use for screening the best designs for experimental validation eventually?

[buerbaumer]

Hi @bee8617,

I don’t think BoltzGen currently has a dedicated dsRNA-binder protocol/module, but the code does support RNA and RNA-protein complexes, so I would treat dsRNA as a structured nucleic-acid target rather than a special mode.

Practically, I would start from the DNA-binder style YAML, but use a .cif/.pdb containing the dsRNA duplex as the target. The main YAML difference is just that the target chains should be RNA chains, and the relevant RNA residues can be marked under binding_types if you want to bias the interface:

entities:
  - protein:
      id: P
      sequence: 100..250

  - file:
      path: dsrna_target.cif
      include:
        - chain:
            id: A
        - chain:
            id: B
      structure_groups: "all"
      binding_types:
        - chain:
            id: A
            binding: 1..30
        - chain:
            id: B
            binding: 1..30

I would use the regular protein-anything pipeline first, then rank by the usual complex metrics: high design_to_target_iptm, low min_interaction_pae / interface PAE, good refolded complex geometry, interface buried surface area, and reasonable noncovalent contacts.

For the specificity requirement, I would not rely on BoltzGen alone. I’d counter-screen the same designed proteins against matched dsDNA and ssRNA structures using the folding/refolding step or an external structure predictor, then reject designs that also show high-confidence binding to those negative targets. For sequence non-specific dsRNA binding, I’d also test the candidates across several unrelated dsRNA duplex sequences and keep designs whose interface remains high-confidence across the panel.

Regards,
Harald (Contributor)