Metagenomics & Microbiome Analysis

Integrated pipeline for exploring microbiome studies, classifying taxa, assessing genome quality, linking microbial composition to clinical phenotypes, and interpreting findings through pathway analysis and literature context.

Guiding principles:

1. Study context first -- understand biome, sequencing method, and metadata before diving into taxa
2. Taxonomic consistency -- GTDB taxonomy as reference standard; reconcile NCBI where needed
3. Genome quality matters -- CheckM completeness/contamination thresholds determine trustworthy MAGs
4. Interpretation over enumeration -- explain what taxa mean for the biological question
5. English-first queries -- use English terms in tool calls

LOOK UP, DON'T GUESS

When uncertain about any scientific fact, SEARCH databases first rather than reasoning from memory.

COMPUTE, DON'T DESCRIBE

When analysis requires computation (statistics, data processing, scoring, enrichment), write and run Python code via Bash. Don't describe what you would do — execute it and report actual results. Use ToolUniverse tools to retrieve data, then Python (pandas, scipy, statsmodels, matplotlib) to analyze it.

Core Databases

Workflow

Phase 0: Parse query → organism, biome, phenotype, or accession
Phase 1: Study Discovery → MGnify_search_studies, ENAPortal_search_studies
Phase 2: Taxonomic Classification → GTDB_search_genomes, GTDB_get_species, GTDB_search_taxon
Phase 3: Genome Quality → MGnify_search_genomes, MGnify_get_genome (CheckM metrics)
Phase 4: Functional Annotation → MGnify GO terms + KEGG pathway mapping
Phase 5: Clinical Associations → GMrepo species-phenotype links
Phase 6: Literature → PubMed/EuropePMC + CTD gene-disease
Phase 7: Interpretation & Report Synthesis

Key Phase Notes

Phase 1: ENA requires structured queries (e.g., study_title="*IBD*"), not free text. If ENA fails, fall back to MGnify.

Phase 2: GTDB uses its own naming (e.g., s__Bacteroides_A fragilis vs NCBI Bacteroides fragilis). Always note discrepancies. Use GTDB_search_taxon(operation="search_taxon", query=name).

Phase 3 - Quality tiers (MIMAG):

• High: >= 90% complete, <= 5% contamination, rRNA + >= 18 tRNAs
• Medium: >= 50% complete, <= 10% contamination
• Low: below medium -- flag but don't exclude

Phase 4 - Functional interpretation: Don't just list GO terms. Connect to biology:

Use kegg_search_pathway(keyword=...) (NOT query). Pathway IDs need organism prefix (hsa, ko, eco), NOT bare map.

Phase 5: GMrepo uses MeSH terms: "Crohn Disease" not "IBD", "Colitis, Ulcerative" not "UC", "Colorectal Neoplasms" not "colorectal cancer". Try NCBI taxon IDs if species name fails.

Phase 6 - Evidence grading:

• Strong: Meta-analysis or >5 studies, consistent direction
• Moderate: 2-5 studies consistent, or 1 large cohort
• Preliminary: Single study or conflicting
• Mechanistic only: In vitro/animal, no human epidemiology

Phase 7 - Report: Executive summary, study landscape, GTDB taxonomy, functional interpretation (not GO term lists), clinical relevance with evidence grades, mechanistic model, genome catalog with quality tiers, data gaps.

Edge Cases & Fallbacks

• Taxon not in GTDB: Try partial search or fall back to MGnify (NCBI taxonomy)
• No GMrepo data: Normal for non-gut organisms; use literature
• GMrepo 0 results: Use formal MeSH terms or NCBI taxon IDs
• No KEGG match: Check MetaCyc or literature

Limitations

• GMrepo: Gut-only
• GTDB: Bacteria/Archaea only
• ENA: Raw data only, strict query syntax
• No sequence analysis: Queries databases, not raw FASTQ/FASTA