What problem does it solve?
Biomedical factual questions — such as which gene belongs to a gene set, which gene is associated with a disease in DisGeNet, or which variant is pathogenic per ClinVar — have authoritative answers in public databases, but language models frequently hallucinate these niche annotations when answering from memory. This Skill grounds every answer in a live database query instead of guessing.
Core Features & Use Cases
- Database-grounded MCQ answering: Routes questions to the correct ToolUniverse tool (MSigDB, DisGeNET, OMIM, MGI, Ensembl, ClinVar, ChEMBL, UniProt, OpenTargets, GTRD, miRDB) and checks each answer option against the returned authoritative data.
- Differential lookups: Handles 'in database X but not database Y' questions by combining UMLS concept resolution, OMIM gene maps, DisGeNET curated associations, and PubTator3 text-mined fallbacks.
- Deterministic computation: Runs Python snippets (Biopython) for exactly computable answers such as ORF counts, restriction digest fragments, and Mendelian segregation ratios, with explicit final-answer-to-option mapping.
- Use Case: Given the question 'Which gene is a predicted target of MIR186-3p according to miRDB?', the Skill queries MSigDB collection C3:MIR:MIRDB with set name MIR186_3P and selects the option present in the returned gene list.
Quick Start
Ask a factual biomedical question such as 'Which of these genes is associated with breast cancer according to DisGeNet but not OMIM?' and let the agent query the appropriate ToolUniverse database tools before answering.
