tooluniverse-model-organism-genetics

## 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.

# Model Organism Genetics Pipeline

Map human genes to model organism orthologs and retrieve phenotype, expression, and functional data across six species. Synthesize cross-species evidence to assess gene function conservation and identify the best animal models for studying human genes and diseases.

**Not for**: human variant interpretation (`tooluniverse-variant-analysis`), drug target validation (`tooluniverse-drug-target-validation`), human disease characterization (`tooluniverse-multiomic-disease-characterization`).

**LOOK UP, DON'T GUESS**: When asked about a species' taxonomy, ecology, or biology, search GBIF/NCBI Taxonomy first. For GBIF: use `GBIF_search_species(query="species name")`, then use the `nubKey` (not `key`) from the result to call `GBIF_get_species(speciesKey=nubKey)` for full taxonomy (kingdom, phylum, class, order, family). The `nubKey` is the GBIF backbone key; the `key` is dataset-specific and often lacks higher taxonomy.

---

## Reasoning Principles

### Ortholog Reasoning
Sequence conservation across species implies functional conservation — but not always. A highly conserved gene in mouse and human likely has the same function. But regulatory differences (when/where a gene is expressed) can cause different phenotypes even from the same gene. Always check: is the protein domain conserved, or just raw sequence? Are there known regulatory differences? A 40% identity ortholog with a conserved catalytic domain can be more functionally equivalent than a 90% identity paralog in the same species.

Paralog contamination is a common pitfall. Gene families (e.g., FOXP1/2/3/4, HOX clusters) generate false ortholog hits. Distinguish true orthologs from paralogs by checking synteny (conserved gene neighborhood) and homology type: 1:1 = likely true ortholog; 1:many or many:many = likely paralog expansion. If the target species has a single gene where humans have multiple (e.g., one fly FoxP vs four human FOXPs), it is the co-ortholog of all human paralogs — note this explicitly.

### Model Organism Selection
Choose your model by the question:
- **Mouse**: mammalian physiology, drug testing, immune system, CNS disease — best when you need human-like biology
- **Fly**: genetic screens, signaling pathways (Notch, Wnt, Hh first characterized here), neural circuits, aging — best for rapid genome-wide genetics
- **Worm**: cell lineage, apoptosis, RNAi screens, aging — best when you need single-cell resolution and mapped connectome
- **Zebrafish**: development, organ formation, live imaging, cardiac biology — best when you need vertebrate biology with optical access
- **Yeast**: cell cycle, DNA repair, metabolism, protein trafficking, chromatin — best for fundamental cell biology
- **Frog (Xenopus)**: early development, cell signaling, oocyte biochemistry — note X. laevis is allotetraploid (two homeologs: .L and .S)

Invertebrates (fly, worm, yeast) lack adaptive immunity and many vertebrate-specific organs — if the question involves those systems, they will be uninformative.

### Phenotype Transfer Reasoning
A knockout phenotype in mouse does not automatically predict the human phenotype. Ask three questions before inferring cross-species relevance:
1. **Is the pathway conserved?** A mouse cardiac phenotype only predicts human cardiac disease if the same developmental pathway operates in both hearts.
2. **Are there compensating paralogs?** If the mouse has one gene but humans have three paralogs, a mouse knockout can be more severe than loss of a single human paralog. Conversely, if humans lost a paralog that mice retain, the mouse KO may overpredict human phenotype.
3. **Is the gene dosage-sensitive?** Haploinsufficiency in mouse (heterozygous phenotype) is a stronger predictor of human dominant disease than phenotypes seen only in homozygous knockouts.

When phenotypes differ across species, consider regulatory divergence: the coding sequence may be conserved while the expression pattern has shifted. This can produce organisms with the "same gene" but different tissues of expression and therefore different phenotypes.

---

## Pipeline

### Phase 0: Human Gene Disambiguation (ALWAYS FIRST)

1. `MyGene_query_genes(query="<gene>")` — get Ensembl ID, Entrez ID, UniProt, symbol (filter by `symbol` match; first hit may be a pseudogene)
2. `ensembl_lookup_gene(gene_id="<ensembl_id>", species="homo_sapiens")` — validate
3. If disease context: `HPO_search_terms(query="<disease>")` — get HPO terms for phenotype matching

Fallback if gene not found: `UniProt_search(query="<gene>", organism="9606")`

**Output**: canonical symbol, Ensembl ID (ENSG), Entrez ID, UniProt accession.

---

### Phase 1: Ortholog Mapping

**Primary**: `EnsemblCompara_get_orthologues(gene="<ENSG>", species="human", target_species="<species>")`

Accepted `target_species` values: `"mouse"`, `"zebrafish"`, `"drosophila_melanogaster"` (NOT "fruitfly" — returns HTTP 400), `"caenorhabditis_elegans"`, `"saccharomyces_cerevisiae"`, `"xenopus_tropicalis"`

**Fallbacks** (if Ensembl Compara returns no results):
1. `PANTHER_ortholog(gene_id="<symbol>", organism=9606, target_organism=<taxon>)` — taxon IDs: mouse=10090, fly=7227, worm=6239, zebrafish=7955, yeast=559292, frog=8364
2. `NCBIDatasets_get_orthologs(gene_id="<entrez_id>")` — broad, all vertebrates
3. For fly: `FlyMine_search(query="<human_gene_symbol>")` — text search finds distant orthologs that automated tools miss; confirm with `FlyBase_get_gene_orthologs`
4. For worm: `WormBase_get_gene(gene_id="<gene_symbol>")` — gene record often contains ortholog info

**Cross-reference via Monarch**:
- `Monarch_search_gene(query="<gene_symbol>")