Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →mims-harvard/tooluniverse-biomedical-fact-lookup
Answer biomedical FACTUAL / recall / multiple-choice questions by querying ToolUniverse database tools instead of answering from memory.
Passed 16 of 20 test cases with the skill, and 16 of 20 without it.
3 fixed15 passed either way1 still failing3 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 3 test cases the skill fixed. Same model, same prompt, with and without the skill.
I have the following single-stranded DNA sequence: 'ATGCGTACGTTAGCTAGCTAGCTGATCGATCGATCGATCGATCGATCGATCGATCGATCGATCGATCGATCGATCG'. How many open reading frames across this sequence string produce a translation product strictly exceeding 15 amino acids before reaching an in-frame stop codon? Options: A) 0 B) 1 C) 2 D) 3 Please format your response by providing the final multiple-choice letter (A-D), the exact calculated integer count, and a brief note on the reading frames and strand orientation evaluated.
| Checked | Without skill | With skill |
|---|---|---|
| Evaluates open reading frames strictly on the forward single strand. | Passed | Passed |
| Reports the exact calculated integer count for ORFs with translation products exceeding 15 amino acids. | Missed | Passed |
The instructions your model reads when the skill loads.
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.
Each was observed producing a wrong answer on a real question. Check them before answering; the detail for each is further down.
"Highest p-value" in GWAS means most significant — the smallest number.
Read literally it picks the study's weakest hit (rs2476491 at 1e-06 instead
of rs7775055-G at 3e-174).
A window of "N bp upstream plus M bp downstream" spans N+M+1 bases — the anchor counts. 100 either side of a TSS is 201 nt, not 200. Check the length you got against the length you asked for.
HPA subcellular locations pool every cell line the antibody was tested in.
Report both main_locations and additional_locations, but when the question
names a line, treat them as candidates and drop annotations belonging to
another line — reciting all five is as wrong as reciting one.
Allen Brain: answer the leaf structure, not its parent. The atlas colours
the specific structure and gives the parent a different colour, so
"Hypothalamus" is wrong where Lateral preoptic area (#F2483B) is right.
AllenBrain_search_structures returns color_hex_triplet.
SCREEN's is_proximal is unreliable; filter on element_type — PLS
and pELS are TSS-proximal, dELS distal.
Derived scores are release-pinned. gnomAD pLI for APOC2 is 0.047 in r4 and 0.402 in r2.1 — an 8.5x difference for the same gene. Set the release the question names and say which you used.
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 " 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.
test cases passed, vs 16 / 20 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.