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:
- Structure-first - Generate backbone geometry before sequence
- Target-guided - Design binders with target structure in mind
- Iterative validation - Predict structure to validate designs
- Developability-aware - Consider aggregation, immunogenicity, expression
- Evidence-separated - Report folding, pose, geometry, protonation and experimental evidence independently
- Actionable output - Provide sequences ready for experimental testing
- 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
For iterative campaigns, conflicting predictions, glycosylated receptor checks,
conditional pH binding, or competition preparation, use
Protein Design Campaign
alongside this workflow. Verify installed tool schemas before executing examples;
model confidence and successful execution do not establish binding.
Report-First Approach (MANDATORY)
- Create
[TARGET]_protein_design_report.md first with section headers
- Progressively update as designs are generated
- Output
[TARGET]_designed_sequences.fasta and [TARGET]_top_candidates.csv
Design Documentation (MANDATORY)
Every design MUST include: Sequence, Length, Target, Method, and available Quality Metrics (pLDDT, pTM, MPNN score, binding prediction). Mark unavailable metrics unmeasured or uncomputed; do not fabricate binding predictions or experimental values.
NVIDIA NIM Tools
| Tool |
Purpose |
Key Parameter |
NvidiaNIM_rfdiffusion (requires NVIDIA_API_KEY env var; free key at build.nvidia.com) |
Backbone generation |
contigs + input_pdb (both required), diffusion_steps (NOT num_steps; default 15) |
NvidiaNIM_proteinmpnn (requires NVIDIA_API_KEY env var; free key at build.nvidia.com) |
Sequence design |
input_pdb (required; NOT pdb or pdb_string), num_seq_per_target, sampling_temp (a list) |
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 (plus required contigs and input_pdb) |
NvidiaNIM_proteinmpnn (requires NVIDIA_API_KEY) |
pdb=content or pdb_string=content |
input_pdb=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 |
Folding and Developability Screening
These are illustrative computational screening tiers, not binding-evidence grades
or calibrated success probabilities. Benchmark screening with appropriate controls.
Monomer pLDDT/pTM cannot establish a complex pose, affinity, cross-species binding
or pH selectivity. Report those dimensions separately, including missing evidence.
| Tier |
Criteria |
| T1 (strongest in this screen) |
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
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
1---23name: tooluniverse-protein-therapeutic-design-24description: "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."5metadata:6 source: https://skillmd.com/skills/mims-harvard/tooluniverse-protein-therapeutic-design-27---89# Therapeutic Protein Designer1011AI-guided de novo protein design using RFdiffusion backbone generation, ProteinMPNN sequence optimization, and structure validation for therapeutic protein development.1213**KEY PRINCIPLES**:141. **Structure-first** - Generate backbone geometry before sequence152. **Target-guided** - Design binders with target structure in mind163. **Iterative validation** - Predict structure to validate designs174. **Developability-aware** - Consider aggregation, immunogenicity, expression185. **Evidence-separated** - Report folding, pose, geometry, protonation and experimental evidence independently196. **Actionable output** - Provide sequences ready for experimental testing207. **English-first queries** - Always use English terms in tool calls2122Therapeutic 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.2324## LOOK UP, DON'T GUESS25When uncertain about any scientific fact, SEARCH databases first rather than reasoning from memory. A database-verified answer is always more reliable than a guess.2627---2829## COMPUTE, DON'T DESCRIBE30When 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.3132## When to Use3334Apply when user asks to:35- Design a protein binder, therapeutic protein, or scaffold36- Optimize a protein sequence for function37- Design a de novo enzyme38- Generate protein variants for target binding3940---4142## Workflow Overview4344```45Phase 1: Target Characterization46 Get structure (PDB, EMDB cryo-EM, AlphaFold), identify binding epitope4748Phase 2: Backbone Generation (RFdiffusion)49 Define constraints, generate >= 5 backbones, filter by geometry5051Phase 3: Sequence Design (ProteinMPNN)52 Design >= 8 sequences per backbone, sample with temperature control5354Phase 4: Structure Validation (ESMFold/AlphaFold2)55 Predict structure, compare to backbone, assess pLDDT/pTM5657Phase 5: Developability Assessment58 Aggregation, pI, expression prediction5960Phase 6: Report Synthesis61 Ranked candidates, FASTA, experimental recommendations62```6364---6566## Critical Requirements6768For iterative campaigns, conflicting predictions, glycosylated receptor checks,69conditional pH binding, or competition preparation, use70[Protein Design Campaign](../tooluniverse-protein-design-campaign/SKILL.md)71alongside this workflow. Verify installed tool schemas before executing examples;72model confidence and successful execution do not establish binding.7374### Report-First Approach (MANDATORY)751. Create `[TARGET]_protein_design_report.md` first with section headers762. Progressively update as designs are generated773. Output `[TARGET]_designed_sequences.fasta` and `[TARGET]_top_candidates.csv`7879### Design Documentation (MANDATORY)80Every design MUST include: Sequence, Length, Target, Method, and available Quality Metrics (pLDDT, pTM, MPNN score, binding prediction). Mark unavailable metrics unmeasured or uncomputed; do not fabricate binding predictions or experimental values.8182---8384## NVIDIA NIM Tools8586| Tool | Purpose | Key Parameter |87|------|---------|---------------|88| `NvidiaNIM_rfdiffusion` *(requires NVIDIA_API_KEY env var; free key at build.nvidia.com)* | Backbone generation | `contigs` + `input_pdb` (both required), `diffusion_steps` (NOT `num_steps`; default 15) |89| `NvidiaNIM_proteinmpnn` *(requires NVIDIA_API_KEY env var; free key at build.nvidia.com)* | Sequence design | `input_pdb` (required; NOT `pdb` or `pdb_string`), `num_seq_per_target`, `sampling_temp` (a list) |90| `ESMFold_predict_structure` | Fast validation | `sequence` (NOT `seq`) |91| `NvidiaNIM_alphafold2` *(requires NVIDIA_API_KEY env var; free key at build.nvidia.com)* | High-accuracy structure inference from sequence | `sequence`, `algorithm` |92| `NvidiaNIM_esm2_650m` *(requires NVIDIA_API_KEY env var; free key at build.nvidia.com)* | Sequence embeddings | `sequences`, `format` |9394### Common Parameter Mistakes9596| Tool | Wrong | Correct |97|------|-------|---------|98| `NvidiaNIM_rfdiffusion` *(requires NVIDIA_API_KEY)* | `num_steps=50` | `diffusion_steps=50` (plus required `contigs` and `input_pdb`) |99| `NvidiaNIM_proteinmpnn` *(requires NVIDIA_API_KEY)* | `pdb=content` or `pdb_string=content` | `input_pdb=content` |100| `ESMFold_predict_structure` | `seq="MVLS..."` | `sequence="MVLS..."` |101| `NvidiaNIM_alphafold2` *(requires NVIDIA_API_KEY)* | `seq="MVLS..."` | `sequence="MVLS..."` |102103### NVIDIA NIM Requirements104- **API Key**: `NVIDIA_API_KEY` environment variable required105- **Rate limits**: 40 RPM (1.5 second minimum between calls)106- AlphaFold2 may return 202 (polling required); RFdiffusion and ESMFold are synchronous107108---109110## Supporting Tools111112| Tool | Purpose | Key Parameters |113|------|---------|----------------|114| `PDBe_get_uniprot_mappings` | Find PDB structures | `uniprot_id` |115| `RCSBData_get_entry` | Download PDB file | `pdb_id` |116| `alphafold_get_prediction` | Get AlphaFold DB structure | `accession` |117| `EMDB_search_structures` | Search cryo-EM maps | `query` |118| `EMDB_get_structure` | Get entry details | `entry_id` |119| `UniProt_get_entry_by_accession` | Get target sequence | `accession` |120| `InterPro_get_protein_domains` | Get domains | `accession` |121122---123124## Folding and Developability Screening125126These are illustrative computational screening tiers, not binding-evidence grades127or calibrated success probabilities. Benchmark screening with appropriate controls.128Monomer pLDDT/pTM cannot establish a complex pose, affinity, cross-species binding129or pH selectivity. Report those dimensions separately, including missing evidence.130131| Tier | Criteria |132|------|----------|133| T1 (strongest in this screen) | pLDDT >85, pTM >0.8, low aggregation, neutral pI |134| T2 | pLDDT >75, pTM >0.7, acceptable developability |135| T3 | pLDDT >70, pTM >0.65, developability concerns |136| T4 | Failed validation or major developability issues |137138---139140## Completeness Checklist141142- [ ] Target structure obtained (PDB or predicted)143- [ ] Binding epitope identified144- [ ] >= 5 backbones generated, top 3-5 selected145- [ ] >= 8 sequences per backbone, MPNN scores reported146- [ ] All sequences validated (ESMFold), pLDDT/pTM reported, >= 3 passing147- [ ] Developability assessed (aggregation, pI, expression)148- [ ] Ranked candidate list, FASTA file, experimental recommendations149150---151152## Reference Files153154- **DESIGN_PROCEDURES.md** - Phase-by-phase code examples, sampling parameters, fallback chains155- **TOOLS_REFERENCE.md** - Complete tool documentation with code examples156- **EXAMPLES.md** - Sample design workflows and outputs157- **CHECKLIST.md** - Detailed phase checklists and quality metrics158- **design_templates.md** - Report templates and output format examples