ProteinMPNN (Claude Skill)

Designs amino-acid sequences that fold to a target backbone using ProteinMPNN, driven as a Claude skill.

   
Type Claude Skill
Supplier Baker Lab, UW Institute for Protein Design
Availability GA — Claude Science research skill
Pricing Free / OSS (MIT)
Capabilities Read/Write — local inference; writes designed sequences (FASTA) with per-position scores
Verified works · 2026-07-20
Security cleared · 2026-07-20 — provenance matches dauparas/ProteinMPNN, MIT, no advisories

How to install

  • Claude Science — enable the built-in ProteinMPNN research skill (Anthropic-hosted; not published to the public anthropics/life-sciences marketplace).
  • Run the model yourself — the upstream model is open source:
    git clone https://github.com/dauparas/ProteinMPNN
    

    Follow the repo README for environment setup and model weights.

What it does

Runs ProteinMPNN to generate sequences for a fixed protein backbone (PDB input), with control over fixed/tied positions, temperature, and symmetry. Local inference only; commonly paired with RFdiffusion backbones and AlphaFold validation.

Primary use cases: De novo protein sequence design, backbone redesign, binder-sequence generation

Notes

Claude Science: Featured as a research skill in Anthropic’s Claude Science. Its inclusion there is an independent signal of quality and trustworthiness for life-science work.

Local (your compute). For ligand/metal/nucleotide context use LigandMPNN; for solubility-tuned designs use SolubleMPNN.

Sources


Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.