| name | chembl-search |
| description | Query the ChEMBL database for bioactive molecules, drug targets, bioactivity measurements (IC50/Ki/EC50), approved drugs, and mechanisms. Use when the user asks about compounds, targets, potency values, or drug mechanisms. Public REST API, no credential needed. |
| license | Apache-2.0 |
ChEMBL Database Query (Codex-native)
Codex-native adaptation of Google DeepMind's chembl-database skill
(science-skills, Apache-2.0).
Uses the public ChEMBL REST API directly through Codex's own tools.
Gates (ask before proceeding)
- Terms notice: on first use in a workspace, tell the user to review the
ChEMBL terms
(https://chembl.gitbook.io/chembl-interface-documentation/about) and record the
notice in provenance.
- Network / install: read-only public HTTP. Be polite about rate limits.
Write responses to a file and parse them; if you write helper code, run it via
uv run, never bare python3.
Workflow
Search molecules first with science_search_chembl; use the direct API below
for target, activity, mechanism, and indication records.
- Choose the endpoint on
https://www.ebi.ac.uk/chembl/api/data/ (append
?format=json):
molecule — compounds by ChEMBL ID, name, or structure.target — drug targets (link to a UniProt accession via
target_components).activity — bioactivity measurements for a molecule/target pair.mechanism / drug_indication — mechanism of action and indications.
- Resolve identifiers first (e.g. compound name → ChEMBL ID, gene/protein →
target via
$cx-uniprot-search then target search) before pulling
activities.
- Run the query, writing JSON to a file rather than dumping large payloads
to stdout; parse with
jq or a short script. Results are usually wrapped in a
list keyed by the endpoint (e.g. molecules, activities).
- Interpret bioactivity correctly (this is where errors happen):
- Report
standard_type (IC50, Ki, EC50, …), standard_value, and
standard_units together — a value is meaningless without its type/unit.
- Prefer
pchembl_value for cross-assay comparison; still note it is
assay-dependent.
- Do not average or compare across different assays, targets, or organisms as
if equivalent; check
assay_type/confidence and target.
- Report compounds/targets with ChEMBL IDs, the requested measurements with
units, and links; separate strong evidence from single low-confidence assays.
- Provenance & review: record queries, endpoints, and cited ChEMBL IDs with
$science-provenance; check claims with $science-review.
Boundaries
- Cite only ChEMBL IDs and values actually returned; never invent IDs, potencies,
or units.
- Bioactivity data are heterogeneous and assay-dependent; state confidence and
never present them as clinical or dosing guidance.
- Approved-drug/indication data describe the record, not medical advice.
