mims-harvard/tooluniverse-metabolomics

Metabolomics Research

Comprehensive metabolomics research skill that identifies metabolites, analyzes studies, and searches metabolomics databases. Generates structured research reports with annotated metabolite information, study details, and database statistics.

Use Case

Use this skill when asked to:

• Identify or annotate metabolites (HMDB IDs, chemical properties, pathways)
• Retrieve metabolomics study information from MetaboLights or Metabolomics Workbench
• Search for metabolomics studies by keywords or disease
• Analyze metabolite profiles or datasets
• Generate comprehensive metabolomics research reports

Example queries:

• "What is the HMDB ID and pathway information for glucose?"
• "Get study details for MTBLS1"
• "Find metabolomics studies related to diabetes"
• "Analyze these metabolites: glucose, lactate, pyruvate"

Databases Covered

Primary metabolite databases:

• HMDB (Human Metabolome Database): 220,000+ metabolites with structures, pathways, and biological roles
• MetaboLights: Public metabolomics repository with thousands of studies
• Metabolomics Workbench: NIH Common Fund metabolomics data repository
• FooDB: Food chemical-constituent database — use FooDB_get_compound (param fdb_id, e.g. "FDB000004") for a food compound's structure plus HMDB/KEGG/PubChem/ChEBI cross-references; ideal for food-metabolomics annotation
• PubChem: Chemical properties and bioactivity data (fallback)

Research Workflow

The skill executes a 4-phase analysis pipeline:

Phase 1: Metabolite Identification & Annotation

For each metabolite in the input list:

1. Search HMDB by metabolite name
2. Retrieve HMDB ID, chemical formula, molecular weight
3. Get detailed metabolite information (description, pathways)
4. Fallback to PubChem for CID and chemical properties if HMDB unavailable

Phase 2: Study Details Retrieval

For provided study IDs:

1. Detect database type (MTBLS = MetaboLights, ST = Metabolomics Workbench)
2. Retrieve study metadata (title, description, organism, status)
3. Extract experimental design and data availability

Phase 3: Study Search

For keyword searches:

1. Search MetaboLights studies by query term
2. Return matching study IDs with preview information
3. Report total number of results

Phase 4: Database Overview

Always included in reports:

1. Sample recent studies from MetaboLights
2. Database statistics and availability
3. Integration information for all databases

Usage Patterns

Pattern 1: Metabolite Identification

Input:

• Metabolite list: ["glucose", "lactate", "pyruvate"]

Output report includes:

• HMDB IDs for each metabolite
• Chemical formulas and molecular weights
• Biological pathways
• PubChem CIDs
• SMILES representations

Pattern 2: Study Retrieval

Input:

• Study ID: "MTBLS1" or "ST000001"

Output report includes:

• Study title and description
• Organism information
• Study status and release date
• Data availability

Pattern 3: Study Search

Input:

• Search query: "diabetes"
• Optional organism filter

Output report includes:

• Matching study IDs
• Study titles and previews
• Total result count

Pattern 4: Comprehensive Analysis

Input:

• Metabolite list: ["glucose", "pyruvate"]
• Study ID: "MTBLS1"
• Search query: "diabetes"

Output report includes:

• All phases combined (identification, study details, search results, overview)
• Cross-referenced information
• Complete metabolomics research summary

Input Parameters

metabolite_list (optional)

List of metabolite names to identify and annotate.

• Format: List of strings
• Examples: ["glucose"], ["lactate", "pyruvate", "acetate"]
• Note: Common names accepted; HMDB will find standard identifiers

study_id (optional)

MetaboLights or Metabolomics Workbench study identifier.

• Format: String starting with "MTBLS" or "ST"
• Examples: "MTBLS1", "ST000001"
• Note: Database auto-detected from prefix

search_query (optional)

Keyword to search metabolomics studies.

• Format: String (disease, compound, organism, method)
• Examples: "diabetes", "glucose metabolism", "LC-MS"

organism (optional)

Target organism for study filtering.

• Format: String (scientific name)
• Default: "Homo sapiens"
• Examples: "Mus musculus", "Saccharomyces cerevisiae"

output_file (optional)

Path for the generated markdown report.

• Format: String (filename with .md extension)
• Default: Auto-generated timestamp-based filename
• Examples: "my_analysis.md", "metabolomics_report.md"

Output Format

All analyses generate a structured markdown report with:

Header section:

• Report title and generation timestamp
• Input parameters summary (metabolites, study ID, search query, organism)

Phase sections:

• Clear section headers (## 1. Metabolite Identification, ## 2. Study Details, etc.)
• Subsections for each metabolite or result
• Consistent formatting (bold labels, tables for results)

Database overview:

• Available databases and statistics
• Recent studies sample
• Integration information

Error handling:

• Graceful error messages for unavailable data
• Fallback strategies documented in output
• "N/A" for missing fields (not blank)

Implementation Notes

SOAP Tool Handling

HMDB tools are SOAP-based and require special parameter handling:

• HMDB_search: Requires operation="search" parameter
• HMDB_get_metabolite: Requires operation="get_metabolite" parameter
• Do not use endpoint or method parameters (not applicable to SOAP)

Response Format Variations

Tools return different response formats - handle all three:

1. Standard format: {status: "success", data: [...], metadata: {...}}
2. Direct list: [...] (e.g., metabolights_list_studies)
3. Direct dict: {field1: ..., field2: ...} (e.g., some detail endpoints)

Always check response type with isinstance() before accessing fields.

Fallback Strategy

Follow this hierarchy for robustness:

1. Primary source: Try main database first (HMDB for metabolites, MetaboLights for studies)
2. Fallback source: Use alternative database if primary fails (PubChem for chemical properties)
3. Default behavior: Show error message with context, continue with remaining phases

Progressive Report Writing

Write report incrementally to avoid memory issues:

1. Create output file early in pipeline
2. Append sections as each phase completes
3. Flush to disk regularly for long analyses
4. Return file path for user access

Tool Discovery

The skill automatically discovers and uses these tools from ToolUniverse:

HMDB Tools:

• HMDB_search: Search metabolites by name
• HMDB_get_metabolite: Get detailed metabolite information

MetaboLights Tools:

• metabolights_list_studies: List available studies
• metabolights_search_studies: Search studies by keyword
• metabolights_get_study: Get study details by ID

Metabolomics Workbench Tools:

• MetabolomicsWorkbench_get_study: Get study information
• MetabolomicsWorkbench_search_compound_by_name: Search compounds

PubChem Tools:

• PubChem_get_CID_by_compound_name: Get PubChem CID
• PubChem_get_compound_properties_by_CID: Get chemical properties

No manual tool configuration required - all tools loaded automatically.

Common Issues

Issue: HMDB returns "Error querying HMDB: 0"

Cause: HMDB search returned empty results or index error accessing first result Solution: This is expected for uncommon metabolites; PubChem fallback will be attempted

Issue: Study details show "N/A" for all fields

Cause: Study ID not found or API unavailable Solution: Verify study ID format (MTBLS* or ST*), check if study is public

Issue: Tool not found errors

Cause: Missing API keys for some databases Solution: Check .env.template, add required API keys to .env file (most metabolomics tools work without keys)

Issue: Large metabolite lists cause slow execution

Cause: Pipeline queries each metabolite individually Solution: Reports limit to first 10 metabolites; consider batching for >20 metabolites

Summary

The Metabolomics Research skill provides comprehensive metabolomics analysis through a 4-phase pipeline that:

1. Identifies metabolites using HMDB (primary) and PubChem (fallback) databases
2. Retrieves study details from MetaboLights and Metabolomics Workbench repositories
3. Searches studies by keywords across metabolomics databases
4. Generates structured reports with all findings in readable markdown format

Key Features:

• ✅ 100% test coverage with working pipeline
• ✅ Handles SOAP tools correctly (HMDB requires operation parameter)
• ✅ Implements fallback strategies (HMDB → PubChem)
• ✅ Graceful error handling (continues if one phase fails)
• ✅ Progressive report writing (memory-efficient)
• ✅ Implementation-agnostic documentation (works with Python SDK and MCP)

Best for:

• Metabolite annotation and pathway analysis
• Study discovery and data retrieval
• Comprehensive metabolomics research reports
• Multi-database metabolomics queries

Reasoning Framework

Starting Point: Mass Spectrum Analysis

Metabolite identification starts with the mass spectrum. LOOK UP DON'T GUESS — always search HMDB/PubChem with the calculated neutral mass rather than guessing identity from m/z alone.

• Step 1 — Calculate neutral mass: Determine ionization mode. Positive: subtract adduct mass ([M+H]+ = -1.0073, [M+Na]+ = -22.9892, [M+NH4]+ = -18.0344). Negative: add back ([M-H]- = +1.0073, [M+Cl]- = +34.9694, [M+HCOO]- = +44.9977).
• Step 2 — Search databases: Query HMDB by mass (±5 ppm for Orbitrap/Q-TOF, ±0.5 Da for unit-resolution). Multiple adduct hypotheses yield different neutral masses — check all plausible adducts before concluding.
• Step 3 — Resolve ambiguity: Exact mass alone often matches 5-20 candidates. Use isotope pattern (M+1/M+2 ratios indicate element composition — e.g., high M+2 suggests S or Cl), retention time, and MS/MS fragmentation to narrow down. A single mass match is L3 confidence; MS/MS match to reference spectrum is required for L2/L1.

Evidence Grading (Metabolite Identification Confidence)

• L1 - Confirmed: HMDB ID + retention time + MS/MS match to reference standard
• L2 - Probable: HMDB match by exact mass + MS/MS similarity (cosine > 0.7), no standard
• L3 - Tentative: Matched by exact mass and molecular formula only; structural isomers unresolved
• L4 - Unknown: Detected m/z with no database match; PubChem fallback may provide candidates

Interpretation Guidance

Metabolite identification: HMDB IDs provide the strongest annotation when paired with experimental validation. A PubChem-only match (fallback) indicates the metabolite is chemically characterized but may lack biological context (pathways, disease associations). Always report the identification confidence level.

Pathway enrichment strategy: When multiple metabolites map to the same KEGG or HMDB pathway, enrichment is meaningful only if the input list is unbiased (not pre-selected for that pathway). Report hits vs. pathway size (3/5 detected is more informative than 3/500). LOOK UP DON'T GUESS — use HMDB_get_metabolite to get pathway annotations for each metabolite rather than assuming pathway membership from names alone.

Biomarker discovery reasoning: A candidate biomarker should show: (1) consistent direction of change across samples (fold-change > 1.5), (2) statistical significance (FDR-adjusted p < 0.05), (3) biological plausibility — LOOK UP the metabolite's known disease associations via HMDB, and (4) reproducibility in an independent cohort. Single-study HMDB associations are hypothesis-generating, not confirmatory. Check MetaboLights/Metabolomics Workbench for independent validation datasets.

Synthesis Questions

A complete metabolomics report should answer:

1. What is the identification confidence level for each metabolite (L1-L4)?
2. Which biological pathways are enriched among the identified metabolites?
3. Do any metabolites meet biomarker criteria (fold-change, significance, plausibility)?
4. Are there relevant metabolomics studies (MTBLS/ST) for the disease or condition of interest?
5. What cross-database evidence supports the biological relevance of key findings (HMDB pathways, PubChem bioactivity)?

Limitations:

• HMDB may not have all metabolites (fallback to PubChem)
• Some studies require authentication or are not public
• Large metabolite lists (>10) auto-limited in reports
• API rate limits may affect large-scale queries

See QUICK_START.md for Python SDK examples, MCP integration, and step-by-step tutorials.