tooluniverse-protein-therapeutic-design

$npx mdskill add mims-harvard/ToolUniverse/tooluniverse-protein-therapeutic-design

Designs therapeutic proteins using AI-guided de novo methods.

  • Designs novel protein binders, scaffolds, and enzyme variants for therapy.
  • Depends on RFdiffusion, ProteinMPNN, and structure validation tools.
  • Uses target structure and developability constraints to guide design.
  • Outputs validated sequences with confidence scores for experimental testing.

SKILL.md

.github/skills/tooluniverse-protein-therapeutic-designView on GitHub ↗
---
name: tooluniverse-protein-therapeutic-design
description: AI-guided de novo protein design — RFdiffusion backbone generation, ProteinMPNN sequence design, structure validation (pLDDT, pTM, MPNN scores). Use for designing therapeutic protein binders, novel scaffolds, enzyme variants, and miniprotein/protein-interface design before experimental validation.
disable-model-invocation: true
---

# Therapeutic Protein Designer

AI-guided de novo protein design using RFdiffusion backbone generation, ProteinMPNN sequence optimization, and structure validation for therapeutic protein development.

**KEY PRINCIPLES**:
1. **Structure-first** - Generate backbone geometry before sequence
2. **Target-guided** - Design binders with target structure in mind
3. **Iterative validation** - Predict structure to validate designs
4. **Developability-aware** - Consider aggregation, immunogenicity, expression
5. **Evidence-graded** - Grade designs by confidence metrics
6. **Actionable output** - Provide sequences ready for experimental testing
7. **English-first queries** - Always use English terms in tool calls

Therapeutic protein design starts with the target interaction. What binding surface do you need to cover? A small pocket = nanobody or peptide. A large flat surface = designed protein. Stability, immunogenicity, and manufacturability constrain the design space.

## LOOK UP, DON'T GUESS
When uncertain about any scientific fact, SEARCH databases first rather than reasoning from memory. A database-verified answer is always more reliable than a guess.

---

## COMPUTE, DON'T DESCRIBE
When analysis requires computation (statistics, data processing, scoring, enrichment), write and run Python code via Bash. Don't describe what you would do — execute it and report actual results. Use ToolUniverse tools to retrieve data, then Python (pandas, scipy, statsmodels, matplotlib) to analyze it.

## When to Use

Apply when user asks to:
- Design a protein binder, therapeutic protein, or scaffold
- Optimize a protein sequence for function
- Design a de novo enzyme
- Generate protein variants for target binding

---

## Workflow Overview

```
Phase 1: Target Characterization
  Get structure (PDB, EMDB cryo-EM, AlphaFold), identify binding epitope

Phase 2: Backbone Generation (RFdiffusion)
  Define constraints, generate >= 5 backbones, filter by geometry

Phase 3: Sequence Design (ProteinMPNN)
  Design >= 8 sequences per backbone, sample with temperature control

Phase 4: Structure Validation (ESMFold/AlphaFold2)
  Predict structure, compare to backbone, assess pLDDT/pTM

Phase 5: Developability Assessment
  Aggregation, pI, expression prediction

Phase 6: Report Synthesis
  Ranked candidates, FASTA, experimental recommendations
```

---

## Critical Requirements

### Report-First Approach (MANDATORY)
1. Create `[TARGET]_protein_design_report.md` first with section headers
2. Progressively update as designs are generated
3. Output `[TARGET]_designed_sequences.fasta` and `[TARGET]_top_candidates.csv`

### Design Documentation (MANDATORY)
Every design MUST include: Sequence, Length, Target, Method, and Quality Metrics (pLDDT, pTM, MPNN score, binding prediction).

---

## NVIDIA NIM Tools

| Tool | Purpose | Key Parameter |
|------|---------|---------------|
| `NvidiaNIM_rfdiffusion` *(requires NVIDIA_API_KEY env var; free key at build.nvidia.com)* | Backbone generation | `diffusion_steps` (NOT `num_steps`) |
| `NvidiaNIM_proteinmpnn` *(requires NVIDIA_API_KEY env var; free key at build.nvidia.com)* | Sequence design | `pdb_string` (NOT `pdb`) |
| `ESMFold_predict_structure` | Fast validation | `sequence` (NOT `seq`) |
| `NvidiaNIM_alphafold2` *(requires NVIDIA_API_KEY env var; free key at build.nvidia.com)* | High-accuracy structure inference from sequence | `sequence`, `algorithm` |
| `NvidiaNIM_esm2_650m` *(requires NVIDIA_API_KEY env var; free key at build.nvidia.com)* | Sequence embeddings | `sequences`, `format` |

### Common Parameter Mistakes

| Tool | Wrong | Correct |
|------|-------|---------|
| `NvidiaNIM_rfdiffusion` *(requires NVIDIA_API_KEY)* | `num_steps=50` | `diffusion_steps=50` |
| `NvidiaNIM_proteinmpnn` *(requires NVIDIA_API_KEY)* | `pdb=content` | `pdb_string=content` |
| `ESMFold_predict_structure` | `seq="MVLS..."` | `sequence="MVLS..."` |
| `NvidiaNIM_alphafold2` *(requires NVIDIA_API_KEY)* | `seq="MVLS..."` | `sequence="MVLS..."` |

### NVIDIA NIM Requirements
- **API Key**: `NVIDIA_API_KEY` environment variable required
- **Rate limits**: 40 RPM (1.5 second minimum between calls)
- AlphaFold2 may return 202 (polling required); RFdiffusion and ESMFold are synchronous

---

## Supporting Tools

| Tool | Purpose | Key Parameters |
|------|---------|----------------|
| `PDBe_get_uniprot_mappings` | Find PDB structures | `uniprot_id` |
| `RCSBData_get_entry` | Download PDB file | `pdb_id` |
| `alphafold_get_prediction` | Get AlphaFold DB structure | `accession` |
| `EMDB_search_structures` | Search cryo-EM maps | `query` |
| `EMDB_get_structure` | Get entry details | `entry_id` |
| `UniProt_get_entry_by_accession` | Get target sequence | `accession` |
| `InterPro_get_protein_domains` | Get domains | `accession` |

---

## Evidence Grading

| Tier | Criteria |
|------|----------|
| T1 (best) | pLDDT >85, pTM >0.8, low aggregation, neutral pI |
| T2 | pLDDT >75, pTM >0.7, acceptable developability |
| T3 | pLDDT >70, pTM >0.65, developability concerns |
| T4 | Failed validation or major developability issues |

---

## Completeness Checklist

- [ ] Target structure obtained (PDB or predicted)
- [ ] Binding epitope identified
- [ ] >= 5 backbones generated, top 3-5 selected
- [ ] >= 8 sequences per backbone, MPNN scores reported
- [ ] All sequences validated (ESMFold), pLDDT/pTM reported, >= 3 passing
- [ ] Developability assessed (aggregation, pI, expression)
- [ ] Ranked candidate list, FASTA file, experimental recommendations

---

## Reference Files

- **DESIGN_PROCEDURES.md** - Phase-by-phase code examples, sampling parameters, fallback chains
- **TOOLS_REFERENCE.md** - Complete tool documentation with code examples
- **EXAMPLES.md** - Sample design workflows and outputs
- **CHECKLIST.md** - Detailed phase checklists and quality metrics
- **design_templates.md** - Report templates and output format examples

More from mims-harvard/ToolUniverse

SkillDescription
create-tooluniverse-skillCreate high-quality ToolUniverse skills following test-driven, implementation-agnostic methodology.
devtu-auto-discover-apisAutomatically discover life science APIs online, create ToolUniverse tools, validate them, and prepare integration PRs. Performs gap analysis to identify missing tool categories, web searches for APIs, automated tool creation using devtu-create-tool patterns, validation with devtu-fix-tool, and git workflow management. Use when expanding ToolUniverse coverage, adding new API integrations, or systematically discovering scientific resources.
devtu-benchmark-harnessContinuous improvement system for ToolUniverse tools, skills, and plugin. Run benchmarks, diagnose failures, route fixes to devtu skills, retest. Use after skill optimization, tool additions, or as regression check.
devtu-code-optimization>
devtu-create-toolCreate new scientific tools for ToolUniverse framework with proper structure, validation, and testing. Use when users need to add tools to ToolUniverse, implement new API integrations, create tool wrappers for scientific databases/services, expand ToolUniverse capabilities, or follow ToolUniverse contribution guidelines. Supports creating tool classes, JSON configurations, validation, error handling, and test examples.
devtu-docs-qualityTOP PRIORITY skill — find and immediately fix or remove every piece of wrong, outdated, or redundant information in ToolUniverse docs. Wrong code, broken links, incorrect counts, and overlapping instructions must be fixed or removed — never left in place. Runs five phases: (D) static method scan, (C) live code execution, (A) automated validation, (B) ToolUniverse audit, (E) less-is-more simplification. Core philosophy: each concept appears exactly once; remove don't add; no emojis; single setup entry point. Use when reviewing docs, before releases, after API changes, or when asked to audit, fix, or simplify documentation.
devtu-fix-toolFix failing ToolUniverse tools by diagnosing test failures, identifying root causes, implementing fixes, and validating solutions. Use when ToolUniverse tools fail tests, return errors, have schema validation issues, or when asked to debug or fix tools in the ToolUniverse framework.
devtu-githubGitHub workflow for ToolUniverse - push code safely by moving temp files, activating pre-commit hooks, running tests, and cleaning staged files. Use when pushing to GitHub, fixing CI failures, or cleaning up before commits.
devtu-optimize-descriptionsOptimize tool descriptions in ToolUniverse JSON configs for clarity and usability. Reviews descriptions for missing prerequisites, unexpanded abbreviations, unclear parameters, and missing usage guidance. Use when reviewing tool descriptions, improving API documentation, or when user asks to check if tools are easy to understand.
devtu-optimize-skillsOptimize ToolUniverse skills for better report quality, evidence handling, and user experience. Apply patterns like tool verification, foundation data layers, disambiguation-first, evidence grading, quantified completeness, and report-only output. Use when reviewing skills, improving existing skills, or creating new ToolUniverse research skills.