tooluniverse-natural-product-dereplication — Claude Code skill
tooluniverse-natural-product-dereplication determines whether a compound is a known microbial natural product, identifies the producing organism, and assigns its ChemOnt chemical class.
Apache-2.0 120 installs ★ 1,798 updated 2026-10-24
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-natural-product-dereplication -a claude-codeWhat tooluniverse-natural-product-dereplication does
Searches NPAtlas (a database of microbial natural products from bacteria and fungi) for structure matches, retrieves the producing organism and literature reference, assigns the ChemOnt hierarchy (kingdom → superclass → class → subclass) via ClassyFire, and cross-references identity in PubChem. Accepts names, formulas, exact masses, InChIKeys, or SMILES. OPSIN resolves systematic IUPAC names to structures; PubChem supplies InChIKeys for trivial names.
NPAtlas scope is microbial only: bacteria and fungi. A negative result means "not a known microbial NP," not proof of absolute novelty — plant, animal, and marine-invertebrate metabolites fall outside its coverage. LOTUS searches across the full tree of life and suits broader checks. MIBiG surfaces biosynthetic gene clusters for identified producers.
All tools run without API keys. The skill does not cover drug discovery, ADMET analysis (use tooluniverse-chemical-compound-retrieval), or metabolomics pathway enrichment (use tooluniverse-metabolomics series).
What it helps with
- check whether a compound is a known microbial natural product
- identify the producing microorganism and original publication
- assign ChemOnt chemical class to a molecule by InChIKey
- resolve an IUPAC name to structure and obtain an InChIKey
- distinguish a new structure from isomers of known products
Example prompts for the agent
- Is this a known microbial natural product? Formula C15H10N2O2, exact mass 250.074.
- Which microbe produces staurosporine and where was it published?
- Assign the ChemOnt class to molecule
BSYNRYMUTXBXSQ-UHFFFAOYSA-Nand show the hierarchy.
How to install tooluniverse-natural-product-dereplication
1Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-natural-product-dereplication -a claude-code2Manually: put the files in place
~/.claude/skills/tooluniverse-natural-product-dereplication/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/tooluniverse-natural-product-dereplication/
1Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-natural-product-dereplication -a codex2Manually: put the files in place
.agents/skills/tooluniverse-natural-product-dereplication/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-natural-product-dereplication -a cursor2Manually: put the files in place
.cursor/skills/tooluniverse-natural-product-dereplication/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install mims-harvard/ToolUniverse skills/tooluniverse-natural-product-dereplication2Manually: put the files in place
.github/skills/tooluniverse-natural-product-dereplication/SKILL.md1Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-natural-product-dereplication -a gemini-cli2Platform CLI
gemini skills install https://github.com/mims-harvard/ToolUniverse --path skills/tooluniverse-natural-product-dereplicationManually: put the files in place
.devin/skills/tooluniverse-natural-product-dereplication/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-natural-product-dereplication -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about tooluniverse-natural-product-dereplication
How do I install tooluniverse-natural-product-dereplication in Claude Code?
Run npx skills add mims-harvard/ToolUniverse --skill tooluniverse-natural-product-dereplication -a claude-code, then check ~/.claude/skills/tooluniverse-natural-product-dereplication/SKILL.md.
What does tooluniverse-natural-product-dereplication do?
It checks whether a compound is a known microbial natural product, retrieves the producing organism and literature reference via NPAtlas, and assigns the ChemOnt chemical class hierarchy.
Do I need API keys to use this skill?
No. NPAtlas, ClassyFire, OPSIN, PubChem, LOTUS, and MIBiG all operate without keys.
Source text
SKILL.md · Apache-2.0
---
name: tooluniverse-natural-product-dereplication
description: Dereplicate a putative natural product and assign its chemical taxonomy. Use to answer "is [compound] a known natural product", "what microbe/organism produces [compound]", "what chemical class is [compound]", "dereplicate this metabolite (by formula/exact mass/InChIKey/SMILES)", or "classify this molecule into ChemOnt". Searches NPAtlas for known microbial natural products (producing organism + literature reference), assigns the ChemOnt kingdom→superclass→class→subclass hierarchy via ClassyFire, resolves systematic IUPAC names to structure via OPSIN, and cross-references identity in PubChem. NOT for general drug/compound identity or ADMET (use tooluniverse-chemical-compound-retrieval / tooluniverse-small-molecule-discovery) and NOT for metabolomics pathway/enrichment analysis (use tooluniverse-metabolomics skills).
disable-model-invocation: true
---
# Natural Product Dereplication & Chemotaxonomy
Decide whether a putative natural product is **already known**, identify the **microbe that produces it**, attach the **literature reference**, and assign its **ChemOnt chemical class**. This is the dereplication question every NP chemist and metabolomics analyst asks of a new feature: *"have we seen this before, and what makes it?"*
**LOOK UP DON'T GUESS**: Never assume an NPAID, producing organism, exact mass, or chemical class. Every identity, provenance, and taxonomy claim must come from a live tool call.
**Scope (microbial NPs only)**: NPAtlas covers natural products from **bacteria and fungi**. It does NOT cover plant, animal, or marine-invertebrate metabolites unless a microbial producer was reported. A "no NPAtlas hit" therefore means *not a known microbial NP* — it does not prove the molecule is novel in an absolute sense.
---
## Backing Tools (all keyless; verify before quoting)
| Tool | Input | Returns |
|------|-------|---------|
| `NPAtlas_search_compounds` | `name` / `inchikey` / `formula` / `smiles`, `limit` | list of {npaid, name, molecular_formula, molecular_weight, exact_mass, inchikey, smiles}. (origin_organism is `null` here — fetch the full record for provenance) |
| `NPAtlas_get_compound` | `npaid` (e.g. `NPA014588`) | full record incl. `origin_organism` (producing microbe + taxonomic lineage) and `origin_reference` (title/doi/journal/year) |
| `ClassyFire_classify_by_inchikey` | `inchikey` (full 27-char) | ChemOnt kingdom→superclass→class→subclass→direct_parent, molecular_framework, substituents. `classified:false` if not in cache |
| `OPSIN_name_to_structure` | `name` (systematic IUPAC) | smiles / inchi / inchikey. `parsed:false` for trade/trivial names |
| `PubChem_get_CID_by_compound_name` | `name` | `{IdentifierList:{CID:[...]}}` |
| `PubChem_get_compound_properties_by_CID` | `cid`, `properties` (e.g. `["MolecularFormula","MolecularWeight","InChIKey","IUPACName"]`) | property table — use to obtain an InChIKey for arbitrary compounds |
| `LOTUS_search_natural_products` | `query` (name / SMILES / InChIKey) | list of natural-product occurrence records (`lotus_id`, `smiles`, `inchikey`, `contains_sugar`, `heavy_atom_number`) |
| `LOTUS_search_by_inchikey` | `inchikey` | same shape as above, InChIKey-scoped |
| `LOTUS_get_compound` | `compound_id` (internal LOTUS id) | full occurrence record incl. reported source organism(s) |
| `MIBiG_list_clusters` | `search_term` (compound/organism name) | biosynthetic gene clusters (BGC accession, quality, completeness, product name(s)) |
| `MIBiG_get_stats` | none | repository-wide counts by status and biosynthetic class (PKS/NRPS/terpene/...) |
**LOTUS vs. NPAtlas scope**: NPAtlas is microbial-only (bacteria + fungi). **LOTUS covers the full tree of life** — plant, animal, marine-invertebrate, and microbial natural products — so a "no NPAtlas hit" that you want to interpret more broadly (is this a known natural product from *any* organism, not just a microbe?) should also be checked against LOTUS before calling something non-natural or novel. LOTUS's own record is an occurrence report (structure + reported source), not a curated single "origin organism" the way NPAtlas's full record is — expect multiple/looser source attributions.
**MIBiG as a producer follow-up**: once NPAtlas (or LOTUS) identifies a producing organism for a microbial natural product, `MIBiG_list_clusters` (search by compound or organism name) can surface the actual experimentally-characterized biosynthetic gene cluster responsible, if one has been deposited — useful for going from "this microbe makes it" to "here is the gene cluster that makes it." `MIBiG_get_stats` gives repository-wide context (e.g. how many clusters exist per biosynthetic class) with no query needed.
---
## Workflow
```
Phase 0: Classify input — name / formula / exact mass / InChIKey / SMILES?
Phase 1: Obtain an InChIKey (the universal key for ClassyFire & precise NPAtlas match)
Phase 2: Dereplicate against NPAtlas (known microbial NP? which organism? which paper?)
Phase 3: Assign ChemOnt chemical class via ClassyFire
Phase 4: Cross-reference identity in PubChem
Phase 5: Report — known/novel call + provenance + class hierarchy + interpretation note
```
### Phase 0 — Classify the input
- **Full InChIKey** (27 chars, `XXXXXXXXXXXXXX-XXXXXXXXXX-X`) → skip to Phase 2; it is already the universal key.
- **Molecular formula / exact mass** → go straight to NPAtlas formula search (Phase 2); these are the rawest dereplication inputs (typical of an untargeted MS feature).
- **SMILES** → usable directly in `NPAtlas_search_compounds(smiles=...)`; also feed to PubChem for an InChIKey.
- **Systematic IUPAC name** (e.g. `2-acetyloxybenzoic acid`) → Phase 1 via OPSIN.
- **Trivial / trade / common name** (e.g. `staurosporine`, `penicillin`) → Phase 1 via PubChem (OPSIN will return `parsed:false` for these).
### Phase 1 — Obtain an InChIKey
```python
# Systematic IUPAC name → structure (OPSIN). parsed:false ⇒ fall through to PubChem.
op = tu.tools.OPSIN_name_to_structure(name="2-acetyloxybenzoic acid")
inchikey = op["data"]["inchikey"] # only if op["data"]["parsed"]
# Trivial/common name → PubChem CID → properties (incl. InChIKey)
cid = tu.tools.PubChem_get_CID_by_compound_name(name="staurosporine")["data"]["IdentifierList"]["CID"][0]
props = tu.tools.PubChem_get_compound_properties_by_CID(
cid=cid, properties=["MolecularFormula","MolecularWeight","InChIKey","IUPACName"])
inchikey = props["data"]["PropertyTable"]["Properties"][0]["InChIKey"]
```
The InChIKey is what makes dereplication exact: an InChIKey match is a structure match; a name match is not (synonyms, analogs, and salts share names).
### Phase 2 — Dereplicate against NPAtlas
Search by the most specific key available. Prefer **InChIKey** (exact structure), then **formula** (catches isomers — useful for an MS feature with only a formula), then **name** (loosest — returns analogs).
```python
# Exact, structure-level
hits = tu.tools.NPAtlas_search_compounds(inchikey="HKSZLNNOFSGOKW-FYTWVXJKSA-N", limit=5)
# MS-feature style (formula or exact mass) — expect multiple isomeric hits
hits = tu.tools.NPAtlas_search_compounds(formula="C28H26N4O3", limit=10)
```
For each candidate NPAID, fetch the **full record** to get the producing organism and reference (search results carry `origin_organism: null`):
```python
rec = tu.tools.NPAtlas_get_compound(npaid="NPA014588")["data"]
organism = rec["origin_organism"]["name"] # e.g. "Streptomyces"
lineage = rec["origin_organism"]["ancestors"] # domain→...→family
reference = rec["origin_reference"] # title, doi, journal, year
```
### Phase 3 — Assign ChemOnt chemical class
```python
cf = tu.tools.ClassyFire_classify_by_inchikey(inchikey=inchikey)["data"]
# cf["kingdom"], cf["superclass"], cf["class"], cf["subclass"], cf["direct_parent"]
# cf["molecular_framework"], cf["substituents"]
```
If `classified:false`, the InChIKey is not in the ClassyFire cache — report the class as *unavailable* (do not invent one). A correct InChIKey is required; a wrong stereo/protonation layer will miss the cache.
### Phase 4 — Cross-reference identity in PubChem
Confirm the same molecule exists in PubChem (CID, IUPAC name, formula, MW) so the identity is anchored to a second independent database. Disagreement in molecular formula between NPAtlas and PubChem is a red flag that the name/structure resolution went astray.
### Phase 5 — Report
Deliver:
1. **Dereplication call** — *Known microbial NP* (with NPAID) **or** *No NPAtlas match (possibly novel / non-microbial)*.
2. **Provenance** — producing organism + taxonomic lineage + literature reference (title, DOI, year).
3. **Chemical class** — ChemOnt kingdom → superclass → class → subclass → direct_parent (+ molecular framework).
4. **Identity cross-refs** — PubChem CID/IUPAC, InChIKey, formula, exact mass.
5. **Interpretation note** (see below).
---
## Interpretation Guidance (dereplication logic)
- **InChIKey hit in NPAtlas** = structure-level match ⇒ confidently a known microbial NP. Quote the NPAID, organism, and paper.
- **Formula / exact-mass hit(s) only** = candidate(s) at the molecular-formula level. Multiple isomers can share one formula and exact mass (e.g. `C28H26N4O3`, 466.2005 Da, returns *both* staurosporine and an ardeemin derivative). Treat these as a **ranked candidate list, not an identification** — confirm with InChIKey, MS/MS, or NMR before claiming identity.
- **Name hit only** = weakest evidence. A name search returns analogs and congeners (searching "staurosporine" returns 5'-hydroxystaurosporine, etc.), not necessarily the exact molecule. Always escalate to an InChIKey check.
- **No NPAtlas hit** = not a known *microbial* natural product. Possible meanings: genuinely novel; a plant/animal/marine metabolite outside NPAtlas scope; or a synthetic/derivatized compound. State which interpretations remain open — do **not** declare "novel" unconditionally.
- **ClassyFire `classified:false`** = ChemOnt has no cached classification for that exact InChIKey (often because the InChIKey's stereo/charge layer differs from the cached entry, or the compound is new). Report class as unavailable rather than guessing.
- **OPSIN `parsed:false`** = the name was not systematic IUPAC (trade/trivial name); route to PubChem for an InChIKey instead.
---
## Worked Example — Staurosporine (trivial name in)
Input: `staurosporine` (a trivial name).
1. **OPSIN** `name=staurosporine` → `parsed:false` (not systematic IUPAC) → fall through to PubChem.
2. **PubChem** name→CID = `44259`; properties → `MolecularFormula C28H26N4O3`, `MW 466.5`, `InChIKey HKSZLNNOFSGOKW-FYTWVXJKSA-N`.
3. **NPAtlas** `inchikey=HKSZLNNOFSGOKW-FYTWVXJKSA-N` → **1 exact hit**: `NPA014588 Staurosporine`, exact_mass `466.2005`. → **Known microbial NP**.
4. **NPAtlas** `get_compound NPA014588` → producing organism **Streptomyces** (genus; lineage Bacteria → Actinobacteria → Actinobacteria → Streptomycetales → Streptomycetaceae); reference *"X-Ray crystal structure of staurosporine: a new alkaloid from a Streptomyces…"*, DOI `10.1039/C39780000800`, 1978.
5. **ClassyFire** `inchikey=HKSZLNNOFSGOKW-FYTWVXJKSA-N` → Organic compounds → **Organoheterocyclic compounds** → Indoles and derivatives → Carbazoles → direct parent **Indolocarbazoles**; molecular framework: aromatic heteropolycyclic.
**Call**: Known microbial natural product (NPA014588), an indolocarbazole alkaloid produced by *Streptomyces*, first reported 1978.
**Dereplication-logic footnote**: the formula `C28H26N4O3` (exact mass 466.2005) alone is *not* unique — NPAtlas formula search returns **2** isomers (staurosporine **and** 5-N-acetyl-15b-didehydroardeemin). The InChIKey is what pins the identity to staurosporine specifically. An MS feature with only this formula would need MS/MS or NMR to choose between the isomers.
## Worked Example — Acetylsalicylic acid (systematic name in)
Input IUPAC `2-acetyloxybenzoic acid`.
1. **OPSIN** → `parsed:true`, `InChIKey BSYNRYMUTXBXSQ-UHFFFAOYSA-N`.
2. **NPAtlas** `inchikey=BSYNRYMUTXBXSQ-...` → no microbial NP record ⇒ *not a known microbial natural product* (it is a semisynthetic drug — consistent with NPAtlas scope).
3. **ClassyFire** → Organic compounds → **Benzenoids** → Benzene and substituted derivatives → Benzoic acids and derivatives → direct parent **Acylsalicylic acids**.
**Call**: Classified into ChemOnt (acylsalicylic acid), but no NPAtlas microbial-NP provenance — illustrating a legitimate "no hit" that is *not* a novel NP.
---
## Limitations (be honest in every report)
- **NPAtlas = microbial NPs only** (bacteria + fungi). No plant, animal, or marine-invertebrate metabolites unless a microbial producer was reported. Absence is not proof of novelty. Check **LOTUS** for broader tree-of-life coverage before concluding non-microbial absence means "not a natural product at all."
- **MIBiG cluster search is name-matched**, not structure-matched — a compound-name hit doesn't guarantee the returned BGC is the correct/only one producing that exact molecule; confirm the product name(s) in the returned record match.
- **ClassyFire is a cache lookup** by InChIKey. `classified:false` means "not cached for this exact InChIKey", not "unclassifiable". Wrong stereo/charge layers miss the cache.
- **Name search ≠ structure match.** NPAtlas/PubChem name searches return synonyms, analogs, and salts. Always confirm identity at the InChIKey level before declaring a compound "known".
- **Formula/exact-mass hits are candidate lists**, not identifications — isomers share formulas. Confirm with orthogonal evidence (MS/MS, NMR, InChIKey).
- **OPSIN parses systematic IUPAC names only** — trivial/trade names return `parsed:false`; route them through PubChem.
- Search-result records carry `origin_organism: null`; provenance requires the `NPAtlas_get_compound` full record.
## Out of Scope → route elsewhere
- General drug/compound **identity & ADMET**, properties, bioassays → **tooluniverse-chemical-compound-retrieval** / **tooluniverse-small-molecule-discovery**.
- Metabolomics **pathway/enrichment** analysis → **tooluniverse-metabolomics** skills.
## Completeness Checklist
- [ ] Input type classified (name / formula / mass / InChIKey / SMILES).
- [ ] An InChIKey was obtained (OPSIN or PubChem) when starting from a name/SMILES.
- [ ] NPAtlas searched with the most specific key available; full record fetched for any hit (organism + reference).
- [ ] Known/novel call made with the correct caveats (microbial-only scope).
- [ ] ChemOnt class reported from ClassyFire, or marked unavailable if `classified:false`.
- [ ] Identity cross-referenced in PubChem (CID/formula/InChIKey agree).
- [ ] Interpretation note included (InChIKey vs formula vs name match; meaning of no-hit / classified:false).
- [ ] Limitations stated; every NPAID, organism, DOI, and class came from a live tool call.
© mims-harvard · Apache-2.0 · mims-harvard/ToolUniverse