How do I install the Zero Shot Mutation Workflow skill?

Run npx skillmds@latest add ai4protein/zero-shot-mutation-workflow in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Zero Shot Mutation Workflow safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Zero Shot Mutation Workflow?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Zero Shot Mutation Workflow free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under Apache-2.

Who published Zero Shot Mutation Workflow?

ai4protein (@ai4protein) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Zero Shot Mutation Workflow by ai4protein

Name: Zero Shot Mutation Workflow
Author: ai4protein

Zero Shot Mutation Workflow

Zero-shot mutation engineering with VenusFactory PLMs. Use when the user wants beneficial mutations, directed evolution candidates, or stability/fitness ranking from a FASTA sequence or PDB structure. Do NOT use for ProteinMPNN inverse folding (proteinmpnn_design_workflow), sequence homology search (protein_sequence_similarity_search), or experimental wet-lab protocols alone.

ai4protein 0ff78d1 3.2 KB Updated Sep 17, 2026

File contents

Zero-Shot Mutation Workflow

Overview

Orchestrates VenusFactory mutation tools: sequence-only PLMs (ESM-1v / ESM2 / VenusPLM) or structure-aware PLMs (ESM-IF1 / SaProt / ProtSSN / MIF-ST). Outputs CSV + heatmap paths in the rich JSON envelope — treat scores as computational hypotheses, not wet-lab results.

VenusFactory execution

read_skill with skill_id: zero_shot_mutation_workflow (this file).
Call the hub tools below by exact name (do not invent Forge/ESM SDK calls).
Parse status, file_info.file_path, and any data / heatmap paths.

Project Tools (VenusFactory2)

Tool	Args	Returns	When
zero_shot_mutation_sequence_prediction	`sequence` or `fasta_file`; `model_name` (default `ESM2-650M`); `backend` (`local`/`pjlab`); optional `out_dir`	status JSON + CSV/heatmap	Only sequence available
zero_shot_mutation_structure_prediction	`structure_file` (PDB); `model_name` (default `ESM-IF1`); `backend`; optional `out_dir`	status JSON + CSV/heatmap	PDB available (prefer over sequence-only)
read_fasta	`file_path`	sequence content	Inspect uploaded FASTA
get_seq_from_pdb_chain_a	`pdb_file`	chain A sequence	Need sequence from structure
render_protein_structure	`pdb_path`, `out_dir`, style options	image under `file_info`	Visualize top mutation sites

Sequence models: ESM-1v, ESM2-650M, ESM-1b, VenusPLM.
Structure models: ESM-IF1, SaProt, ProtSSN, MIF-ST (and related names exposed by the tool).

Workflow

A. Sequence-only path

Ensure FASTA via read_fasta or pass sequence directly.
Call zero_shot_mutation_sequence_prediction with an explicit out_dir under the session workspace.
Summarize top-ranked substitutions; warn that scores are model-dependent.

B. Structure path (preferred when PDB exists)

Validate PDB path (may come from predict_structure_esmfold or download_alphafold_structure_by_uniprot_id).
Call zero_shot_mutation_structure_prediction.
Optional: render_protein_structure highlighting candidate sites; optional cross-check with predict_residue_function (Activity/Binding/Conserved Site).

When NOT to use

Designing a new sequence for a fixed backbone → proteinmpnn_design_workflow
Need experimental structure QA / AlphaFold confidence → alphafold_database / protein_structure_pipeline
Homolog discovery → protein_sequence_similarity_search or foldseek_structural_similarity

Common mistakes

Calling non-existent ESM Forge APIs instead of hub tools.
Using structure models without a real PDB path.
Presenting zero-shot ranks as measured ΔΔG / activity without caveats.
Omitting out_dir when the session needs a stable artifact path.

---name: zero-shot-mutation-workflowdescription: Zero-shot mutation engineering with VenusFactory PLMs. Use when the user wants beneficial mutations, directed evolution candidates, or stability/fitness ranking from a FASTA sequence or PDB structure. Do NOT use for ProteinMPNN inverse folding (proteinmpnn_design_workflow), sequence homology search (protein_sequence_similarity_search), or experimental wet-lab protocols alone.license: Apache-2.0---# Zero-Shot Mutation Workflow## OverviewOrchestrates VenusFactory mutation tools: sequence-only PLMs (ESM-1v / ESM2 / VenusPLM) or structure-aware PLMs (ESM-IF1 / SaProt / ProtSSN / MIF-ST). Outputs CSV + heatmap paths in the rich JSON envelope — treat scores as **computational hypotheses**, not wet-lab results.## VenusFactory execution1. `read_skill` with `skill_id: zero_shot_mutation_workflow` (this file).2. Call the hub tools below by **exact name** (do not invent Forge/ESM SDK calls).3. Parse `status`, `file_info.file_path`, and any `data` / heatmap paths.## Project Tools (VenusFactory2)| Tool | Args | Returns | When ||------|------|---------|------|| **zero_shot_mutation_sequence_prediction** | `sequence` **or** `fasta_file`; `model_name` (default `ESM2-650M`); `backend` (`local`/`pjlab`); optional `out_dir` | status JSON + CSV/heatmap | Only sequence available || **zero_shot_mutation_structure_prediction** | `structure_file` (PDB); `model_name` (default `ESM-IF1`); `backend`; optional `out_dir` | status JSON + CSV/heatmap | PDB available (prefer over sequence-only) || **read_fasta** | `file_path` | sequence content | Inspect uploaded FASTA || **get_seq_from_pdb_chain_a** | `pdb_file` | chain A sequence | Need sequence from structure || **render_protein_structure** | `pdb_path`, **`out_dir`**, style options | image under `file_info` | Visualize top mutation sites |Sequence models: `ESM-1v`, `ESM2-650M`, `ESM-1b`, `VenusPLM`. Structure models: `ESM-IF1`, `SaProt`, `ProtSSN`, `MIF-ST` (and related names exposed by the tool).## Workflow### A. Sequence-only path1. Ensure FASTA via `read_fasta` or pass `sequence` directly.2. Call `zero_shot_mutation_sequence_prediction` with an explicit `out_dir` under the session workspace.3. Summarize top-ranked substitutions; warn that scores are model-dependent.### B. Structure path (preferred when PDB exists)1. Validate PDB path (may come from `predict_structure_esmfold` or `download_alphafold_structure_by_uniprot_id`).2. Call `zero_shot_mutation_structure_prediction`.3. Optional: `render_protein_structure` highlighting candidate sites; optional cross-check with `predict_residue_function` (Activity/Binding/Conserved Site).## When NOT to use- Designing a new sequence for a fixed backbone → `proteinmpnn_design_workflow`- Need experimental structure QA / AlphaFold confidence → `alphafold_database` / `protein_structure_pipeline`- Homolog discovery → `protein_sequence_similarity_search` or `foldseek_structural_similarity`## Common mistakes- Calling non-existent ESM Forge APIs instead of hub tools.- Using structure models without a real PDB path.- Presenting zero-shot ranks as measured ΔΔG / activity without caveats.- Omitting `out_dir` when the session needs a stable artifact path.

Frequently asked questions