Skip to main content
skill CX Codex

tooluniverse-vaccine-design — Claude Code skill: setup guide

tooluniverse-vaccine-design designs computational peptide vaccine candidates using IEDB epitope prediction and HLA coverage. Fits automated workflows for immunogenicity screening.

Apache-2.0 229 installs ★ 1,798 updated 2026-10-24

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-vaccine-design -a claude-code
Open on GitHub

What tooluniverse-vaccine-design does

This skill sets up a systematic workflow for designing subunit and peptide vaccine candidates. It connects computational models for MHC-I and MHC-II binding, linear B-cell epitope detection, and experimental validation via IEDB endpoints. Target antigen sequences are fetched from UniProt or BVBRC, focusing on surface-exposed proteins.

The pipeline covers multiple stages: antigen selection, T-cell epitope ranking via NetMHCpan EL, antibody epitope identification, HLA population coverage, and cross-strain conservation checks. Strong binders are isolated using percentile ranks below 0.5% or experimental IC50 thresholds under 50 nM. Pure computational prediction requires validation against published experimental assays.

The skill explicitly forbids guessing MHC binding or epitope sequences from memory, forcing the model to query tools like IEDB_predict_mhci_binding and iedb_search_epitopes.

What it helps with

  • predict peptide binding to MHC-I and MHC-II complexes
  • identify linear B-cell epitopes from pathogen sequences
  • query experimentally validated assays in the IEDB database
  • evaluate HLA population coverage for vaccine constructs
  • filter surface antigens conserved across pathogen strains

Example prompts for the agent

  • Predict MHC-I binding epitopes for this viral surface glycoprotein across HLA-A*02:01
  • Design a multi-epitope construct targeting conserved regions from UniProt entry P0DTC2
  • Search IEDB for experimentally verified CD4+ T-cell epitopes for my antigen

How to install tooluniverse-vaccine-design

1Skills CLI (Vercel)

npx skills add mims-harvard/ToolUniverse --skill tooluniverse-vaccine-design -a claude-code

2Manually: put the files in place

~/.claude/skills/tooluniverse-vaccine-design/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/tooluniverse-vaccine-design/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about tooluniverse-vaccine-design

How to install tooluniverse-vaccine-design in Claude Code?

Run npx skills add mims-harvard/ToolUniverse --skill tooluniverse-vaccine-design -a claude-code. The definition will be placed in ~/.claude/skills/tooluniverse-vaccine-design/SKILL.md.

What is tooluniverse-vaccine-design used for?

It orchestrates computational candidate vaccine assembly through epitope predictions, conservation checks, and immunogenicity scoring.

Can I use it for antibody engineering tasks?

No, the repository provides a dedicated skill named tooluniverse-antibody-engineering for antibody optimization and engineering.

Source text

SKILL.md · Apache-2.0
---
name: tooluniverse-vaccine-design
description: "Computational vaccine candidate design: peptide/subunit vaccines via MHC-I/MHC-II epitope prediction (IEDB), population HLA coverage optimization, B-cell epitope identification, and cross-strain conservation analysis. Use for vaccine epitope prediction, HLA allele coverage, multi-epitope construct design, and immunogenicity assessment. Combines predicted MHC binding with experimentally validated IEDB epitopes for higher-confidence designs."
disable-model-invocation: true
---

# Vaccine Design

Computational pipeline for designing peptide/subunit vaccine candidates through epitope prediction, population coverage optimization, and immunogenicity assessment.

## Reasoning Strategy

Vaccine design requires presenting the right epitopes to elicit protective immunity — not just any immune response, but one that is neutralizing, durable, and broadly applicable. For T-cell vaccines, the core tool is MHC binding prediction (IEDB tools): predict peptide-MHC affinity across multiple HLA alleles, then select epitopes with broad coverage of the target population. For antibody vaccines, prioritize surface-exposed conserved regions — a deeply buried or hypervariable region makes a poor antibody target. MHC binding does not equal immunogenicity; many good binders are not immunogenic in vivo due to tolerance, poor processing, or lack of T-cell help. A multi-epitope strategy (combining MHC-I for CD8+ CTL response, MHC-II for CD4+ helper response, and B-cell epitopes for antibody induction) is more robust than any single epitope. Conservation across pathogen strains is critical — an epitope that mutates under immune pressure (like HIV envelope hypervariable regions) is a poor vaccine target.

**LOOK UP DON'T GUESS**: Do not predict MHC binding or population coverage from memory — use `IEDB_predict_mhci_binding` / `IEDB_predict_mhcii_binding` for T-cell predictions, `IEDB_predict_bcell_epitopes` for antibody (B-cell) epitope prediction, and `iedb_search_epitopes` for validated experimental data. Do not assume what's on the pathogen surface; retrieve annotated sequences from UniProt or BVBRC.

**Key principles**:
1. **Epitope-driven** — vaccines work by presenting epitopes to T/B cells; start with epitope prediction
2. **Population coverage matters** — HLA diversity means no single epitope covers everyone; design for breadth
3. **Multi-epitope is better** — combine CD8+ (MHC-I) and CD4+ (MHC-II) epitopes for robust immunity
4. **Conservation = broad protection** — conserved epitopes across strains provide cross-protective immunity
5. **Evidence grading** — T1: clinical trial data, T2: in-vivo immunogenicity, T3: in-vitro binding, T4: computational prediction only

---

## When to Use

- "Design a vaccine against [pathogen]"
- "Predict T-cell epitopes for [protein]"
- "What MHC-I epitopes does [protein] have?"
- "Assess population coverage of these epitopes"
- "Find conserved epitopes across [pathogen] strains"

**Not this skill**: For HLA typing or allele frequency only, use `tooluniverse-hla-immunogenomics`. For antibody engineering, use `tooluniverse-antibody-engineering`.

---

## Core Tools

| Tool | Use For |
|------|---------|
| `IEDB_predict_bcell_epitopes` | De-novo predict linear B-cell (antibody) epitopes from sequence (BepiPred/Emini/…) |
| `iedb_search_epitopes` | Search experimentally validated epitopes |
| `iedb_get_epitope_mhc` | Get detailed epitope data (assay results, MHC restriction) |
| `iedb_search_mhc` | Search validated MHC binding assay data |
| `IEDB_predict_mhci_binding` | **Predict MHC-I binding** (NetMHCpan EL; rank < 0.5% = strong binder) |
| `IEDB_predict_mhcii_binding` | **Predict MHC-II binding** (NetMHCIIpan EL; CD4+ helper epitopes) |
| `UniProt_get_entry_by_accession` | Get antigen protein sequence |
| `UniProt_search` | Find pathogen protein sequences |
| `BVBRC_search_genome_features` | Search pathogen proteomes |
| `alphafold_get_prediction` | Get/predict antigen 3D structure |
| `EnsemblVEP_annotate_hgvs` | Check epitope conservation across variants |
| `PubMed_search_articles` | Find published vaccine studies |
| `search_clinical_trials` | Find ongoing vaccine clinical trials |

---

## Workflow

```
Phase 0: Antigen Selection
  Pathogen → essential surface proteins → sequence retrieval
    |
Phase 1: T-Cell Epitope Prediction
  MHC-I (CD8+ CTL) and MHC-II (CD4+ helper) binding prediction
    |
Phase 2: B-Cell Epitope Prediction
  Linear and conformational B-cell epitopes for antibody response
    |
Phase 3: Population Coverage
  HLA allele frequencies → design for target population
    |
Phase 4: Conservation Analysis
  Cross-strain epitope conservation → broad protection
    |
Phase 5: Candidate Assembly & Report
  Multi-epitope construct design → immunogenicity assessment
```

### Phase 0: Antigen Selection

**Best antigens for vaccines**: Surface-exposed, essential for pathogen function, conserved across strains.

```python
# Find pathogen surface proteins
UniProt_search(query="[organism] AND locations:(location:cell surface) AND reviewed:true")
# Or search BVBRC for annotated pathogen proteomes
BVBRC_search_genome_features(keyword="surface protein", genome_id="[taxon_id]")
```

**Antigen prioritization**: prefer surface-exposed (secreted/outer membrane) over cytoplasmic; >95% conserved across strains; essential for pathogen viability; known immunogen in natural infection. Use UniProt subcellular location annotations and PubMed to verify these properties.

### Phase 1: T-Cell Epitope Prediction

**MHC-I epitopes** (CD8+ cytotoxic T cells — kill infected cells):

```python
# Option A: Search for KNOWN validated epitopes from IEDB
iedb_search_mhc(
    mhc_class="I",
    qualitative_measure="Positive",
    filters={"source_organism_iri": "eq.NCBITaxon:2697049"},  # SARS-CoV-2
    select=["linear_sequence", "mhc_restriction", "qualitative_measure"],
    limit=50
)

# Option B: PREDICT novel peptide binding (recommended for new proteins)
IEDB_predict_mhci_binding(
    sequence="YOUR_PROTEIN_SEQUENCE",  # full protein or peptide
    allele="HLA-A*02:01",             # or H-2-Kd for mouse
    method="netmhcpan_el",            # EL = eluted ligand (recommended)
    length=9                           # 8-11 for MHC-I
)
# Returns peptides ranked by percentile_rank:
# < 0.5% = strong binder (include in vaccine)
# 0.5-2% = moderate binder (consider)
# > 2% = weak/non-binder (exclude)
```

**MHC-II epitopes** (CD4+ helper T cells — activate B cells and CD8+ T cells):

```python
iedb_search_mhc(
    mhc_class="II",
    qualitative_measure="Positive",
    filters={"source_organism_iri": "eq.NCBITaxon:2697049"},
    limit=50
)
```

**Binding affinity interpretation**:

| IC50 (nM) | Classification | Vaccine Relevance |
|-----------|---------------|-------------------|
| < 50 | Strong binder | Include — high presentation probability |
| 50-500 | Moderate binder | Consider — may contribute to response |
| 500-5000 | Weak binder | Exclude — unlikely to be presented |
| > 5000 | Non-binder | Exclude |

**HLA supertype strategy**: For broad coverage, predict against HLA supertypes:
- **A2 supertype** (A*02:01, A*02:06, A*68:02) — covers ~40% globally
- **A3 supertype** (A*03:01, A*11:01, A*31:01) — covers ~25%
- **B7 supertype** (B*07:02, B*35:01, B*51:01) — covers ~25%
- **A2 + A3 + B7 + B44** combined — covers >90% of most populations

### Phase 2: B-Cell Epitope Prediction

B-cell epitopes trigger antibody production. Look for:
- **Linear epitopes**: Continuous peptide sequences (easier to synthesize)
- **Conformational epitopes**: 3D surface patches (requires structural data)

```python
# De-novo predict LINEAR B-cell epitopes along the antigen sequence (BepiPred).
# Returns contiguous predicted epitope regions + per-residue scores.
IEDB_predict_bcell_epitopes(sequence="[antigen_aa_sequence]", method="Bepipred")
# Cross-check against KNOWN experimentally validated epitopes
iedb_search_epitopes(query="[protein_name]", epitope_type="B cell")
# Get structure for conformational epitope prediction
alphafold_get_prediction(uniprot_id="[accession]")
```

**B-cell epitope criteria**: Surface-exposed loops, hydrophilic regions, flexible regions (high B-factor). Prefer `IEDB_predict_bcell_epitopes` regions that also fall on surface-exposed loops in the structure; the `method` arg also supports Emini (surface accessibility), Kolaskar-Tongaonkar (antigenicity), and Parker (hydrophilicity).

### Phase 3: Population Coverage

No HLA-frequency tool exists in ToolUniverse, but the coverage math is packaged in
`scripts/population_coverage.py`. Pass the HLA alleles your selected epitopes bind
(from the `IEDB_predict_mhci/mhcii_binding` results) and get the % of the
population covered:

```bash
# Broad first-pass estimate (bundled average frequencies):
python scripts/population_coverage.py --alleles "HLA-A*02:01,HLA-A*01:01,HLA-A*03:01,HLA-A*24:02,HLA-B*07:02,HLA-B*08:01,HLA-B*44:02"
# -> {"overall_coverage": 73.6, "per_locus_coverage_pct": {"A": 62.9, "B": 28.9}, ...}

# Population-SPECIFIC: supply real allele frequencies (ALLELE<TAB>FREQ) for the
# target ethnicity from the Allele Frequency Net Database (allelefrequencies.net)
# or the IEDB population-coverage tool (tools.iedb.org/population):
python scripts/population_coverage.py --alleles-file covered.txt --freq-file afnd_han_chinese.tsv
```
The bundled default is an approximate broad average — do NOT report it as coverage
for a specific ethnicity; use `--freq-file` with AFND/IEDB data for that. Also use
the HLA supertype strategy to ensure your epitope set spans the common supertypes.

**Population coverage targets**:

| Coverage Level | Interpretation | Action |
|---------------|---------------|--------|
| >90% | Excellent — vaccine will work in most individuals | Proceed to development |
| 70-90% | Good — most people covered; some populations underserved | Add more epitopes for uncovered HLA types |
| 50-70% | Moderate — significant gaps | Redesign with broader HLA coverage |
| <50% | Poor — vaccine will miss too many people | Fundamental redesign needed |

### Phase 4: Conservation Analysis

Check if epitopes are conserved across pathogen strains/variants:

```python
# Search for protein variants across strains
PubMed_search_articles(query="[pathogen] [protein] sequence variation strains")
# Check specific mutations in epitope regions
EnsemblVEP_annotate_hgvs(hgvs_notation="[variant_in_epitope]")
```

**Conservation interpretation**:
- **100% conserved** across all known strains → ideal vaccine target
- **>95% conserved** → good target; monitor emerging variants
- **80-95% conserved** → may need strain-specific variants in construct
- **<80% conserved** → avoid; pathogen evolves to escape this epitope

### Phase 5: Candidate Assembly & Report

**Multi-epitope construct design principles**:
1. Include 3-5 MHC-I epitopes (CD8+ response)
2. Include 2-3 MHC-II epitopes (CD4+ helper response)
3. Include 1-2 B-cell epitopes (antibody response)
4. Connect with appropriate linkers (AAY for MHC-I, GPGPG for MHC-II)
5. Add adjuvant sequence if needed (e.g., flagellin domain for TLR5)

**Report structure**:
1. **Antigen Selection** — rationale, conservation, essentiality
2. **Epitope Map** — all predicted epitopes with binding affinities and HLA restrictions
3. **Top Epitopes** — ranked by binding strength × conservation × population coverage
4. **Population Coverage** — % coverage per major world population
5. **Conservation Analysis** — strain coverage, escape risk assessment
6. **Construct Design** — multi-epitope sequence with linkers
7. **Clinical Precedent** — existing vaccines/trials for related antigens
8. **Limitations** — predicted only (T4 evidence); needs experimental validation

---

## Limitations

- **All predictions are computational** (T4 evidence) — experimental validation (binding assays, immunogenicity studies) is required before any clinical development
- **No immunogenicity guarantee** — MHC binding ≠ immunogenicity; many good binders are not immunogenic in vivo
- **B-cell epitope prediction is less reliable** than T-cell; conformational epitopes require accurate structures
- **No adjuvant optimization** — adjuvant selection requires empirical testing
- **Pathogen evasion** — rapidly evolving pathogens (HIV, influenza) may escape epitope-based vaccines

© mims-harvard · Apache-2.0 · mims-harvard/ToolUniverse

Similar skills

All skills for Science and research →

From the same set136

The whole set →
tooluniverse-computational-biophysics tooluniverse-computational-biophysics works through quantitative biology and biophysics problems, from drug clearance calculations to population-genetics comparisons. CX Codex +1 232installs tooluniverse-rare-disease-genomics tooluniverse-rare-disease-genomics is a Claude Code skill for rare-disease research, tracing patient phenotypes to candidate genes and variants and finding clinical trials. CX Codex +1 231installs tooluniverse-cell-line-profiling The tooluniverse-cell-line-profiling skill helps you pick the right cancer cell line for research by matching genes, tumor type, and drug sensitivity across DepMap, Cellosaurus, COSMIC, and PharmacoDB. CX Codex +1 230installs tooluniverse-aging-senescence tooluniverse-aging-senescence structures agent reasoning on aging biology, senescence pathways, and senolytics. It prioritizes causal evidence over correlation across four tiers. CX Codex +1 230installs tooluniverse-variant-functional-annotation tooluniverse-variant-functional-annotation annotates protein variants with ProtVar, ClinVar, gnomAD, CADD, and ClinGen data. Useful for functional context, population evidence, and variant review. CX Codex +1 229installs tooluniverse-stem-cell-organoid tooluniverse-stem-cell-organoid guides stem cell protocols, pluripotency verification, and organoid fidelity scoring using ToolUniverse biomedical tools. CX Codex +1 226installs tooluniverse-noncoding-rna tooluniverse-noncoding-rna guides non-coding RNA queries across miRBase, LNCipedia, and RNAcentral to inspect targets, transcript structures, and disease links. CX Codex +1 225installs tooluniverse-functional-genomics-screens tooluniverse-functional-genomics-screens is a skill for ranking hits from genetic screens (CRISPR-KO, CRISPRi, shRNA) using essentiality, pathway, and clinical data. CX Codex +1 225installs

Project sponsors

Companies that keep this analytics open