BioDatabase Reference Retrieval

Overview

This skill retrieves reference data from two official, well-documented public APIs:

Both are plain HTTP/JSON-or-XML endpoints. No browser automation is required — that is the main difference from the pipeline this skill was extracted from, where every database script drove a Selenium browser session. Scraping a data portal for structured records is strictly worse than asking the API: it breaks whenever a template changes, runs an order of magnitude slower, and produces output that cannot be reproduced from a documented request.

What this skill gets right that a naive implementation does not

CAS numbers come from the right endpoint. The xrefs/RegistryID endpoint does not return CAS numbers. It returns several hundred registry IDs from many different databases mixed together — CHEBI:15365, DB00945, LGCAMP0133.00-11, and the occasional 50-78-2. A CAS number has a fixed format (NNNNN-NN-N), so the reliable way to get one is to filter the synonyms list by that pattern. For CID 2244 (aspirin) the synonyms endpoint yields three CAS candidates; RegistryID yields only two, and neither is guaranteed to be the primary registry number.

Missing CIDs are detectable. PubChem answers a request for a non-existent CID with HTTP 200 and a body of {"CID": 999999999} — just the CID, no properties. Checking the status code alone will therefore report a successful lookup of a compound that does not exist. Every record here carries an explicit found flag.

Lineage includes the taxon itself. NCBI's LineageEx element contains only ancestors. The species level is available solely from the top-level <Rank> and <ScientificName>. Reading only LineageEx leaves the species name — the field you usually want — empty.

All ranks are retained. LineageEx contains dozens of rank values beyond the textbook eight: domain, subphylum, superclass, clade, suborder, infraorder, parvorder, superfamily, subfamily and more. Selecting only the eight standard ranks silently drops domain (Eukaryota / Bacteria / Archaea), which is the single most useful top-level discriminator.

When to Use This Skill

• Getting molecular properties of compounds you already know by CID
• Looking up CAS registry numbers for a set of CIDs
• Converting a list of NCBI Taxonomy IDs into species names and lineages
• Annotating a hits table with compound descriptors or taxonomic classification
• Verifying that a CID actually exists in PubChem

Quick Start

export PYTHONPATH="$PWD/packages"
export BIOPIPE_EMAIL="you@your-institution.edu"   # NCBI 要求的联系邮箱

# 化合物：批量取属性 + 同义词 + CAS
python scripts/fetch_compounds.py --cids cids.txt --output-dir results/

# 化合物：直接给 CID，只要基本属性（不要几百个同义词）
python scripts/fetch_compounds.py --cids 2244 3672 --output-dir results/ --no-synonyms

# 分类学：批量取 lineage
python scripts/fetch_taxonomy.py --tax-ids 9606 10090 --output-dir results/

Workflow

Compounds (fetch_compounds.py)

1. Collect CIDs — --cids accepts CID strings or paths to text files (one per line; # comments and a header row are skipped). Duplicates are removed while preserving order.
2. Fetch properties — one PUG-REST request per 100 CIDs. The default property set covers molecular formula, weight, exact mass, XLogP, TPSA, formal charge, both SMILES variants, InChI, InChIKey, IUPAC name, H-bond donors/acceptors, rotatable bonds and heavy atom count.
3. Fetch synonyms — skipped with --no-synonyms. A well-known compound has several hundred synonyms; the output keeps the first --max-synonyms (default 20) but reports the true synonym_count.
4. Extract CAS numbers — filtered out of the synonyms by CAS format, then topped up from RegistryID for any that the synonyms missed. Skipped entirely with --no-cas.
5. Export — compounds.csv and compounds.json.

Taxonomy (fetch_taxonomy.py)

1. Collect IDs — same file/argument handling as CIDs.
2. Fetch — efetch.fcgi?db=taxonomy&id=...&retmode=xml. Taxonomy IDs are primary keys, so no esearch step is needed (unlike BioProject).
3. Parse — scientific name, common name, rank, and the full lineage.
4. Export — taxonomy.csv (one row per taxon, with domain … species flattened into columns plus a full_lineage string) and taxonomy.json (the full nested record, including every lineage entry).

Parameters

fetch_compounds.py

fetch_taxonomy.py

Outputs

Validation

• Check found before trusting a row. found=False means PubChem has no compound with that CID. This is not a transient error — the identifier is wrong or refers to a compound that was removed.
• A compound may legitimately have several CAS numbers. Salts, hydrates and stereoisomers each get their own. The CAS column holds them semicolon-separated. Match on the primary one only if you are joining against a single-registry table.
• CAS candidates are format-validated, not registry-validated. The pattern check removes non-CAS registry IDs, but it cannot tell you whether a CAS number has been re-assigned. For critical identity work, cross-check the InChIKey.
• superkingdom is empty in current data. NCBI retired that rank; "cellular organisms" is now cellular root and the top-level split is domain (Eukaryota / Bacteria / Archaea). Use the domain column.
• Not every taxon has a species rank. Strain-level entries report no rank (e.g. Escherichia coli str. K-12 substr. MG1655). The rank column tells you what the entry actually is — do not assume species.
• Species names are not stable identifiers. Escherichia coli's phylum was renamed from Proteobacteria to Pseudomonadota; the tax_id is stable, the name is not. Join on tax_id.
• A missing common name is normal. GenbankCommonName is only present for well-studied organisms; most taxa have none.
• Taxonomy IDs are stable, names are not. Never key a cross-study join on the scientific name.

References

• references/methodology.md — API protocols, field semantics, the four PubChem and NCBI traps, CAS extraction logic
• references/troubleshooting.md — common failures and how to diagnose them
• PubChem PUG-REST documentation: https://pubchem.ncbi.nlm.nih.gov/docs/pug-rest
• NCBI Taxonomy: https://www.ncbi.nlm.nih.gov/taxonomy
• NCBI E-utilities documentation: https://www.ncbi.nlm.nih.gov/books/NBK25501/