Alterlab proteinmpnn

Skill AlterLab-IEU/AlterLab-Academic-Skills/skills/bioinformatics/alterlab-proteinmpnn

Design protein sequences for a fixed backbone with ProteinMPNN (Dauparas 2022) — message-passing inverse folding that outputs sequences predicted to fold to a given structure, with fixed positions, tied/symmetric chains, amino-acid bias, and a soluble-model variant. Use when inverse-folding a backbone PDB into sequences, redesigning selected positions, imposing symmetry across chains, or generating the sequence step of a design→fold→score loop. For pocket/interface design WITH a bound ligand, metal, or nucleic acid prefer alterlab-ligandmpnn; to GENERATE a new backbone prefer alterlab-rfdiffusion; to refold and validate a design prefer alterlab-alphafold; for generative multimodal design prefer alterlab-esm. Part of the AlterLab Academic Skills suite.From its SKILL.md

Install

npx -y skills add AlterLab-IEU/AlterLab-Academic-Skills --skill alterlab-proteinmpnn

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

runs commandsInstructs the agent to run 2 commands, including `python parse_multiple_chains.py --input_path=pdbs/ --output_path=parsed.jsonl` and 1 more.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

4.3 KB, 790 tokens by cl100k_base, as published. Nobody here has run it

ProteinMPNN (fixed-backbone sequence design)

Overview

ProteinMPNN (Dauparas et al., Science 2022; dauparas/ProteinMPNN) solves the inverse-folding problem: given a protein backbone (a 3D structure with no or a placeholder sequence), it designs amino-acid sequences predicted to fold to that backbone. It is fast, robust, runs on CPU, and is the standard "sequence" step between backbone generation (alterlab-rfdiffusion) and structure validation (alterlab-alphafold).

When to Use This Skill

Use this skill when the user wants to:

Inverse-fold a backbone PDB into one or more candidate sequences.
Redesign only selected positions while fixing the rest (partial design).
Enforce symmetry by tying residues/chains so homo-oligomers get identical sequences.
Bias the amino-acid composition (e.g. avoid cysteines) or use the soluble model.
Produce the sequence step of a design → fold → score loop.

Does NOT Trigger

Scenario	Use instead
Design a pocket/interface with a ligand, metal, or nucleic acid present	`alterlab-ligandmpnn`
Generate a new backbone (no starting structure)	`alterlab-rfdiffusion`
Refold a designed sequence to check it (validation)	`alterlab-alphafold`
Generative multimodal (sequence+structure+function) design	`alterlab-esm`

Core Capabilities

1. Basic inverse folding

# Parse the PDB, then design sequences (dauparas/ProteinMPNN CLI — TODO(verify) flags/version).
# The parser lives in the repo's helper_scripts directory; run it by name:
python parse_multiple_chains.py --input_path=pdbs/ --output_path=parsed.jsonl
python protein_mpnn_run.py \
  --jsonl_path parsed.jsonl --out_folder out/ \
  --num_seq_per_target 8 --sampling_temp "0.1"

Lower --sampling_temp (e.g. 0.1) gives conservative, high-confidence designs; higher temperatures increase diversity. Output FASTA headers carry the model score (lower = better) and sequence recovery.

2. Fixed positions and chains

Supply a fixed-positions spec (JSONL from the helper scripts) to keep catalytic/known residues while redesigning the rest, and a chain spec to design only some chains. Verify the exact helper-script names against your checkout (TODO(verify)).

3. Symmetry / tied positions

Tie positions across chains so a homo-oligomer receives one sequence applied symmetrically — essential for symmetric alterlab-rfdiffusion outputs.

4. Design → fold → score loop

The canonical de-novo pipeline:

Generate a backbone with alterlab-rfdiffusion.
Design sequences for it here (ProteinMPNN), sampling several per backbone.
Score by refolding each with alterlab-alphafold and accepting only self-consistent designs (returns to the target backbone with high pLDDT, low PAE).

Resources

references/proteinmpnn_usage.md — install/pinning, helper-script inputs (fixed positions, tied chains, bias), the soluble model, temperature guidance, and loop integration. Loaded on demand.

Part of the AlterLab Academic Skills suite.

What ships with it: 2 files

5.1 KB alongside SKILL.md

evals/

evals.json2.9 KB

references/

proteinmpnn_usage.md2.2 KB

Gives 0 of the 12 instructions most design frontend skills give in 790 tokens

Counted across 1,179 of the 2,086 authors here whose files we hold, read 2026-09-06

Commit to a bold aesthetic directionin 31 of 1179, across 24 files
Prefer component composition over inheritancein 28 of 1179, across 14 files
Animate only transform and opacity propertiesin 27 of 1179, across 22 files
Memoize expensive computations with useMemoin 26 of 1179, across 13 files
Use semantic HTML elementsin 24 of 1179, across 23 files
Virtualize long lists for performancein 21 of 1179, across 10 files
Use CSS variables for design tokensin 20 of 1179, across 14 files
Implement loading, empty, and error statesin 20 of 1179
Lazy load heavy components with Suspensein 19 of 1179, across 8 files
Respect prefers-reduced-motion media queriesin 18 of 1179, across 10 files
Prioritize CSS-only animations for HTMLin 18 of 1179, across 16 files
Use compound components for related UI elementsin 18 of 1179, across 7 files

Said here and by no other author read

Parse PDB files into JSONL format
Run protein_mpnn_run.py for sequence design
Use lower sampling temperature for high confidence
Supply fixed positions spec for partial design
Tie positions across chains for symmetric designs
Integrate into design fold score loop

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.