tooluniverse-biomedical-fact-lookup

# Biomedical Fact Lookup (tool-grounded answering)

Factual biomedical questions — "which gene is in set X", "which gene is associated with disease Y according to DisGeNet", "which gene has a TF binding site per GTRD" — have an authoritative answer in a public database. Guessing from memory is unreliable (≈chance on niche annotations); the matching ToolUniverse tool returns the ground truth.

## RULE ZERO: Look it up, never guess

If a question names a database, a gene set, or any annotation that lives in a database, you MUST query the tool before answering. Answering a "according to <database>" question from memory is a failure mode — these annotations (predicted miRNA targets, ChIP-seq binding, curated gene sets, disease associations) are exactly what models hallucinate. A tool-verified answer beats any recalled fact.

## Multiple-choice procedure

Most of these questions are MCQ with an "Insufficient information to answer the question." distractor. Do this:

1. **Parse** the question for: the **named database/collection**, the **anchor entity** (the gene set, disease, miRNA, TF, locus…), and the **candidate options**.
2. **Resolve** the anchor to the right tool + identifier (see Routing table).
3. **Query** the tool once to get the authoritative member list / association set.
4. **Check each option** against that result. Exactly one option should be supported.
5. **Answer** with that option's letter. Only choose "Insufficient information" if the tool genuinely returns nothing for a valid query (not because you skipped the query).

## Routing table — question pattern → tool

| Question mentions… | Tool(s) (verified) | How |
|---|---|---|
| a named **gene set** / **oncogenic signature** (MSigDB C6, e.g. `ATM_DN.V1_DN`) | `MSigDB_get_gene_set_members` | list members, check which option is in it |
| **miRNA target** "according to miRDB" (e.g. MIR186-3p) | `MSigDB_get_gene_set_members` (collection C3:MIR:MIRDB) | set name = `MIR<number>_<3P\|5P>`, e.g. `MIR186_3P` |
| **TF binding site / target** "according to GTRD" (e.g. PGM3) | `MSigDB_check_gene_in_set` (collection C3:TFT:GTRD) | set name = `<TF>_TARGET_GENES`, e.g. `PGM3_TARGET_GENES`; pass `gene` per option |
| **pathway / hallmark** membership | `MSigDB_get_hallmark_geneset`, `MSigDB_get_geneset` | `HALLMARK_<NAME>` or exact set name |
| **gene ↔ disease** association (DisGeNet, OpenTargets, OMIM) | `umls_search_concepts` → `DisGeNET_get_disease_genes`/`DisGeNET_get_gda`; `OpenTargets_*`, `MyDisease_get_disease`, `OMIM_search`; **text-mined fallback:** `PubTator3_LiteratureSearch` / `PubTator3_GetEntityRelations` (`e1=@GENE_<sym>`), `EPMC_get_text_mined_annotations` | DisGeNET needs a **UMLS CUI** (resolve via `umls_search_concepts` → `C0152200`, then `disease=C0152200`) + `DISGENET_API_KEY`. See the "in X but not Y" recipe below |
| **mouse phenotype** gene set (MGI / MP:xxxxx, e.g. "increased carcinoma incidence") | `MGI_search_genes` → `MGI_get_phenotypes` | for **each** candidate gene: search → take the `MGI:` id → `MGI_get_phenotypes`; the matching gene is the one whose `phenotype_statement` list contains the phenotype the question names (see interpretation note) |
| **gene genomic location** (Ensembl band, e.g. chr7q34) | `Ensembl_*` / `NCBIDatasets_get_gene_by_symbol` | resolve each option, compare cytoband/coordinates |
| **variant / sequence** pathogenicity ("which variant/sequence is pathogenic *or* benign per ClinVar") | (only when genuinely unsure) `annotate_variant_multi_source`, `VEP_predict_pathogenicity`, `UniProt_get_disease_variants_by_accession` | **Be efficient — do NOT query every option (that causes timeouts).** Identify the protein once, find each option's single substitution, and reason about the specific residue changes directly; the base model is usually reliable on well-characterized ClinVar variants. Make at most ONE targeted tool call to resolve a truly uncertain variant. **Watch the question's polarity** (benign vs pathogenic): for "most likely benign", a common/reference-matching variant is the answer; for "most likely pathogenic", a rare damaging one is. |
| **drug / compound** target, MoA, approval | `ChEMBL_*`, `OpenFDA_*`, `GtoPdb_*`, `PubChem_*` | resolve drug, query the relation |
| **protein** function / domain / sequence | `UniProt_*` | resolve accession, read annotation |

When unsure which tool wraps a database, search the catalog by the *relation* (e.g. "gene disease association", "gene set members"), not the brand name — ToolUniverse usually already has it.

## MSigDB set-name conventions (the most common LAB-Bench pattern)

ToolUniverse's `MSigDB_*` tools cover several collections that LAB-Bench questions are built from. Get the set name right:

- **C6 oncogenic signatures** — use the exact set name quoted in the question (e.g. `ATM_DN.V1_DN`, `KRAS.600_UP.V1_UP`).
- **C3:MIR:MIRDB** (miRDB v6.0 predicted miRNA targets) — `MIR<number>_<3P|5P>` (e.g. `MIR186_3P`, `MIR675_3P`). This *is* miRDB; do not say "no access to miRDB".
- **C3:TFT:GTRD** (GTRD TF target genes) — `<TF>_TARGET_GENES` (e.g. `PGM3_TARGET_GENES`). This *is* GTRD.
- **Hallmark** — `HALLMARK_<NAME>`.

`MSigDB_get_gene_set_members` (operation `get_gene_set`) returns `{genes:[...]}`; `MSigDB_check_gene_in_set` (operation `check_gene_in_set`, param `gene`) returns `{is_member: bool}`.

## Gene–disease "in database X but NOT database Y" recipe

These questions (e.g. "which gene is associated with disease D according to DisGeNet but **not** OMIM?") need a *differential* lookup, not a single query:

1. Resolve D to a UMLS CUI (`umls_search_concepts`).
2. **OMIM side:** `OMIM_search`/`OMIM_get_gene_map` for D → the set of OMIM-causal genes.
3. **DisGeNet side:** `DisGeNET_get_disease_genes(disease=CUI)` (curated). Note the academic key is **curated-only**; DisGeNet *also* includes a text-mined tier the key can't see.
4. **Text-mined fallback** (covers DisGeNet's text-mined tier when curated is empty): `PubTator3_LiteratureSearch("<