proteinmpnn-design-workflow

name	proteinmpnn_design_workflow
description	ProteinMPNN inverse folding: design or score sequences on a fixed backbone. Use when the user wants sequence design from PDB, interface/binder design, homomer symmetry, or fixed catalytic residues. Do NOT use for zero-shot mutation ranking on a wild-type sequence (zero_shot_mutation_workflow) or de novo fold hallucination without a backbone.
license	Apache-2.0
metadata	{"version":"1.0","skill-author":"VenusFactory2"}

ProteinMPNN Design Workflow

Overview

Orchestrates VenusFactory denovo tools for backbone-conditioned sequence design and scoring. Residue indices in fixed_residues_json are 1-indexed.

Project Tools (VenusFactory2)

Tool	Args	When
proteinmpnn_sequence_design_from_structure	pdb_path; optional designed_chains, fixed_chains, fixed_residues_json, homomer, num_sequences, temperatures, out_dir	Design sequences
proteinmpnn_sequence_scoring_from_structure	pdb_path; optional fasta_path, designed_chains, out_dir	Score designs / native
pdb_chain_sequences / check_pdb_apo	PDB path	Prep / inspect
zero_shot_mutation_structure_prediction	designed PDB context	Optional post-design ranking

Workflow patterns

Goal	Key args
Single-chain redesign	pdb_path only
Design chain B against fixed A	designed_chains=["B"], fixed_chains=["A"]
Keep catalytic residues	fixed_residues_json='{"A":[57,102,195]}'
Homomer	designed_chains=["A","B","C"], homomer=true

1. Validate PDB (check_pdb_apo if ligand-free backbone expected).
2. Design with explicit out_dir.
3. Score designs with proteinmpnn_sequence_scoring_from_structure.
4. Optional: structure prediction / property checks on top sequences.

Common mistakes

• Passing nearly all residues as fixed when intending to redesign the whole chain (or the inverse) — check tool log warnings.
• 0-based residue indices in fixed_residues_json.
• Expecting affinity/ΔG; ProteinMPNN gives sequence likelihoods, not binding free energy.