---

name: disease-structure description: "Use gget for gene-set enrichment, cancer cohort exploration, local COSMIC queries, Open Targets target annotations, and G2P residue/isoform maps; choose identifiers, filters, plots, files, and service/licensing boundaries safely." disable-model-invocation: true metadata: disco-role: operating license: BSD 2-Clause

Disease and structure annotations

Use this sub-skill when the task needs disease, cancer, enrichment, target, residue, isoform, or protein-to-structure annotations from gget. It covers gget.enrichr, gget.cbio_search, gget.cbio_plot, gget.cosmic, gget.opentargets, and gget.g2p. These functions are wrappers around public remote services or user-provided local files; results can change as upstream databases are updated.

Choose the operation

• Gene-set enrichment: start with enrichr. Use a supported shortcut only for human/mouse; use an exact library name for other species. Ensembl IDs require ensembl=True, and an Ensembl background requires ensembl_bkg=True. See Enrichr.
• Cancer cohort discovery/heatmaps: call cbio_search to obtain study IDs, inspect the returned IDs, then call cbio_plot. The latter downloads and caches cBioPortal data and writes a PNG; it does not return a result table. See cBioPortal.
• COSMIC: query a local TSV with an exact, case-insensitive match. Download first only when the user has the required COSMIC account, permission, and storage. See COSMIC.
• Target associations: use opentargets with an Ensembl gene ID and one of its seven resources. Filters are exact equality on returned column names. See Open Targets.
• Residue/isoform annotations: use g2p for G2P features, map, or alignment. Resolve or pass the gene/UniProt pair explicitly; use residue filtering only for per-residue resources. See G2P.

Common execution contract

1. Record the identifier namespace and species before making a request. Keep Ensembl version suffixes when they are meaningful to the upstream service; note that Enrichr conversion strips a suffix and cBio/COSMIC have their own accession matching rules.
2. Run Python calls with verbose=False only after the first diagnostic run; retain the returned object and inspect its type, shape, columns, and a few rows before writing a downstream analysis.
3. Prefer explicit output paths in caller code. gget's save=True conventions differ by module; where a Python function has no save argument, write the returned DataFrame yourself.
4. Treat an empty DataFrame/list, None, a failed plot, a missing cache file, or a service error as an unresolved result—not evidence of no biology. Follow troubleshooting.
5. Do not place credentials in prompts, notebooks, committed code, or command history. COSMIC full downloads require an account and licensing review; public Enrichr, cBioPortal, Open Targets, G2P, and UniProt calls still need network access and may be rate-limited.

Minimal Python recipes

import gget

# Symbols, or Ensembl IDs with ensembl=True.
enrichment = gget.enrichr(["ACE2", "AGT", "AGTR1"], database="ontology")

# Search first; then inspect/select IDs rather than guessing a study ID.
studies = gget.cbio_search(["esophag", "ovarian"])
gget.cbio_plot(studies[:1], ["AKT1", "NOTCH3"], stratification="tissue")

# Existing, licensed/local COSMIC TSV only.
cosmic_rows = gget.cosmic("EGFR", cosmic_tsv_path="path/to/file.tsv", limit=20)

# Target disease associations; use resource-specific filters when narrowing.
diseases = gget.opentargets("ENSG00000169194", resource="diseases", limit=10)

# G2P feature rows at selected residues, with an explicit reviewed accession.
features = gget.g2p("BRCA1", uniprot_id="P38398", residues=[185, 1775, 1812])

Expected Python defaults are pandas DataFrames for Enrichr, COSMIC, Open Targets, and G2P; Enrichr/COSMIC can return JSON-compatible lists with json=True, and cbio_plot returns True after plotting. A remote failure can instead raise, return an empty object, or return None according to the module; never assume one universal error behavior.

Boundaries and routing

• Route raw PDB retrieval, AlphaFold prediction, sequence retrieval, or direct structure download to the sequence-tools sub-skill. G2P map may expose a PDB Ids List, but this sub-skill only teaches the mapping.
• Route mutation sequence transformation and the detailed gget mutate workflow to specialized-workflows. COSMIC's gget_mutate=True export is a handoff, not a mutation-analysis recipe here.
• Open Targets' disease rows are EFO-mapped associations and may be phenotypes or measurements, not only MONDO diseases. G2P's public API does not expose the portal-only gnomAD, ClinVar, or HGMD overlays.
• For full signatures, resource schemas, cache/output names, constraints, and recovery steps, use the linked reference pages rather than guessing from a CLI help screen.

Verification checklist

For a new workflow, verify: (a) the identifier namespace and species are explicit; (b) the selected database/resource/filter names occur in the local reference; (c) the returned type and required columns are checked; (d) output paths are inspected; (e) a small deterministic or mocked case is retained; and (f) network, account, license, cache, and truncation limits are recorded.

Difficult cases to test are an Enrichr Ensembl list with versioned IDs and a small top-N assertion after conversion, plus an Open Targets diseases query whose disease IDs seed a G2P feature request while clearly labeling the portal-only variant-overlay gap. These cases are intentionally more demanding than the original smoke tests.