Tooluniverse Drug Research

Drug Research Strategy

Comprehensive drug investigation using 50+ ToolUniverse tools across chemical databases, clinical trials, adverse events, pharmacogenomics, and literature.

KEY PRINCIPLES:

1. Report-first approach - Create report file FIRST, then populate progressively
2. Compound disambiguation FIRST - Resolve identifiers before research
3. Citation requirements - Every fact must have inline source attribution
4. Evidence grading - Grade claims by evidence strength (T1-T4)
5. Mandatory completeness - All sections must exist, even if "data unavailable"
6. English-first queries - Always use English drug/compound names in tool calls, even if the user writes in another language. Only try original-language terms as a fallback. Respond in the user's language

LOOK UP, DON'T GUESS

When asked about a drug, query ChEMBL/PubChem/DailyMed FIRST. Don't guess at mechanism, targets, or side effects — look them up. When you're not sure about a fact, your first instinct should be to SEARCH for it using tools, not to reason harder from memory.

Drug Mechanism Reasoning

When investigating a drug's mechanism of action, trace the full causal chain:

1. Target engagement - Which protein(s) does the drug bind, and with what affinity/selectivity?
2. Molecular effect - Does binding inhibit, activate, or modulate the target's function?
3. Pathway consequence - Which signaling or metabolic pathway is altered downstream?
4. Cellular phenotype - What changes occur at the cell level (proliferation, apoptosis, secretion)?
5. Physiological outcome - How does the cellular effect translate to the therapeutic benefit in the patient?

Workflow Overview

1. Report-First Approach (MANDATORY)

DO NOT show the search process or tool outputs to the user. Instead:

1. Create the report file FIRST - [DRUG]_drug_report.md with all 11 section headers and [Researching...] placeholders. See REPORT_TEMPLATE.md [blocked] for the full template.
2. Progressively update the report - Replace placeholders with findings as you query each tool.
3. Use ALL relevant tools - Query multiple databases for each data type; cross-reference across sources.

2. Citation Requirements (MANDATORY)

Every piece of information MUST include its source. Use inline citations:

*Source: PubChem via `PubChem_get_compound_properties_by_CID` (CID: 4091)*

3. Progressive Writing Workflow

Step 1:  Create report file with all section headersStep 2:  Resolve compound identifiers -> Update Section 1Step 3:  Query PubChem/ADMET-AI/DailyMed SPL -> Update Section 2 (Chemistry)Step 4:  Query FDA Label MOA + ChEMBL + DGIdb -> Update Section 3 (Mechanism)Step 5:  Query ADMET-AI tools -> Update Section 4 (ADMET)Step 6:  Query ClinicalTrials.gov -> Update Section 5 (Clinical)Step 7:  Query FAERS/DailyMed -> Update Section 6 (Safety)Step 8:  Query PharmGKB -> Update Section 7 (Pharmacogenomics)Step 9:  Query DailyMed/Orange Book -> Update Section 8 (Regulatory)Step 10: Query PubMed/literature -> Update Section 9 (Literature)Step 11: Synthesize findings -> Update Executive Summary & Section 10Step 12: Document all sources -> Update Section 11 (Data Sources)

Compound Disambiguation (Phase 1)

CRITICAL: Establish compound identity before any research.

Identifier Resolution Chain

1. PubChem_get_CID_by_compound_name(compound_name)   -> Extract: CID, canonical SMILES, formula
2. ChEMBL_search_molecules(query=drug_name)   -> Extract: ChEMBL ID, pref_name
3. DailyMed_search_spls(drug_name)   -> Extract: Set ID, NDC codes (if approved)
4. PharmGKB_search_drugs(query=drug_name)   -> Extract: PharmGKB ID (PA...)
5. DrugCentral_search(query=drug_name, size=...)   -> Extract: InChIKey, CAS number, canonical SMILES, cross-referenced FDA/EMA/PMDA approval status in one call

DrugCentral (live-verified: DrugCentral_search, DrugCentral_get_drug, DrugCentral_get_targets) is a useful fifth disambiguation source and a fast Path 2 (Mechanism & Targets) shortcut — DrugCentral_get_targets(chem_id=<InChIKey or ChEMBL ID>) returns target names/classes directly, which can cross-check (not replace) the ChEMBL-activities-derived target table this skill's Path 2 already builds. Real example: DrugCentral_search(query="metformin") returns InChIKey XZWYZXLIPXDOLR-UHFFFAOYSA-N; DrugCentral_get_targets on that InChIKey returns 5'-AMP-activated protein kinase among its targets.

Dietary supplements are not in PubChem/ChEMBL/DailyMed/DrugCentral as prescription drugs — check the NIH DSLD instead. If the query resolves to zero hits in the chain above (or the user is clearly asking about an OTC vitamin/herbal/supplement product, e.g. "vitamin D gummies" or a specific brand), use NIHDSLD_search_products(query=<name or brand>) -> NIHDSLD_get_label(product_id=...) for the full per-serving ingredient list with amounts and %daily value. Live-verified: NIHDSLD_search_products(query="vitamin d") returns real product 20581 ("Vitamin D Gummy Vitamins", Nutrition Now); NIHDSLD_get_label(20581) returns its full label (2000 IU Vitamin D at 500% daily value per 2-gummy serving, plus calories/carbs/sugar) and an off_market flag. This complements rather than replaces the prescription-drug paths above — supplements have no FDA approval/mechanism-of-action data, so most other report sections legitimately stay "not applicable" for a supplement query.

Handle Naming Ambiguity

Research Paths Summary

Each path has detailed tool chains and output examples in REPORT_GUIDELINES.md [blocked].

PATH 1: Chemical Properties & CMC

Tools: PubChem properties -> ADMET-AI physicochemical -> ADMET-AI solubility -> DailyMed chemistry/description Output: Physicochemical table, Lipinski assessment, QED score, salt forms, formulation comparison

PATH 2: Mechanism & Targets

Tools: DailyMed MOA -> ChEMBL activities (NOT ChEMBL_get_molecule_targets) -> ChEMBL target details -> DGIdb -> PubChem bioactivity Critical: Derive targets from activities filtered to pChEMBL >= 6.0. Avoid ChEMBL_get_molecule_targets. Output: FDA MOA text, target table with UniProt/potency, selectivity profile

PATH 3: ADMET Properties

Tools: ADMET-AI (bioavailability, BBB, CYP, clearance, toxicity) Fallback: DailyMed clinical_pharmacology + pharmacokinetics + drug_interactions Critical: If ADMET-AI fails, automatically use fallback. Never leave Section 4 empty.

PATH 4: Clinical Trials

Tools: search_clinical_trials -> compute phase counts -> extract outcomes/AEs -> fda_pharmacogenomic_biomarkers Critical: Section 5.2 must show actual counts by phase/status in table format.

PATH 5: Post-Marketing Safety

Tools: FAERS (reactions, seriousness, outcomes, deaths, age) + DailyMed (DDI, dosing, warnings) Critical: Include FAERS date window, seriousness breakdown, and limitations paragraph.

PATH 6: Pharmacogenomics

Tools: PharmGKB (search -> details -> annotations -> guidelines) Fallback: DailyMed pharmacogenomics section + PubMed literature

PATH 7: Regulatory & Patents

Tools: FDA Orange Book (search, approval history, exclusivity, patents, generics) + DailyMed (special populations via LOINC codes) Note: US-only data; document EMA/PMDA limitation.

PATH 8: Real-World Evidence

Tools: ClinicalTrials.gov (OBSERVATIONAL studies) + PubMed (real-world, registry, surveillance)

PATH 9: Comparative Analysis

Tools: Abbreviated tool chains for each comparator + head-to-head trial search + PubMed meta-analyses

FDA Label Core Fields

For approved drugs, retrieve these DailyMed sections early (after getting set_id):

Fallback Chains

Quick Reference: Tools by Use Case

See TOOLS_REFERENCE.md [blocked] for the complete tool listing with parameters and input format requirements.

Type Normalization

Many tools require string inputs. Always convert IDs before API calls:

• ChEMBL IDs, PubMed IDs, NCT IDs: convert int -> str
• SMILES for ADMET-AI: pass as list ["SMILES_STRING"]
• FAERS drug names: use UPPERCASE (e.g., "METFORMIN")
• ChEMBL IDs: full format "CHEMBL1431" not "1431"
• PharmGKB IDs: PA prefix "PA450657" not "450657"

Common Use Cases

Always maintain all section headers but adjust depth based on query focus and data availability.

When NOT to Use This Skill

• Target research -> Use target-intelligence-gatherer skill
• Disease research -> Use disease-research skill
• Literature-only -> Use literature-deep-research skill
• Single property lookup -> Call tool directly
• Structure similarity search -> Use PubChem_search_compounds_by_similarity directly

Cross-Skill References

For drug interaction checking, run: python3 skills/tooluniverse-drug-drug-interaction/scripts/pharmacology_ref.py --type interaction --drug1 X --drug2 Y

Additional Resources

• Report template: REPORT_TEMPLATE.md [blocked] - Initial file template, citation format, evidence grading, scorecard, audit template
• Report guidelines: REPORT_GUIDELINES.md [blocked] - Detailed section-by-section instructions with output examples
• Tool reference: TOOLS_REFERENCE.md [blocked] - Complete tool listing with parameters and input formats
• Verification checklist: CHECKLIST.md [blocked] - Section-by-section pre-delivery verification
• Examples: EXAMPLES.md [blocked] - Detailed workflow examples for different use cases