К основному содержанию
скилл CX Codex

tooluniverse-gene-disease-association: скилл для Claude Code

Скилл tooluniverse-gene-disease-association связывает гены и заболевания по шести базам данных и рассчитывает оценку согласованности доказательств.

Apache-2.0 240 установок ★ 1 798 обновлён 2026-10-24

Установить в Claude Code · Через Skills CLI (Vercel) Все способы и платформы ↓
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-gene-disease-association -a claude-code
Открыть на GitHub

Что делает tooluniverse-gene-disease-association

Скилл запускает структурированный поиск связей между генами и фенотипами через библиотеку ToolUniverse. Сначала агент резолвит идентификаторы через MyGene и Monarch: сопоставляет символы генов с Ensembl ID, кодами HGNC или номерами MIM, а названия болезней переводит в MONDO или UMLS CUI. Затем данные собираются по слоям: скоринг DisGeNET, мишени OpenTargets, кросс-видовые связи Monarch Initiative и менделевские каталоги OMIM, GenCC, Orphanet и G2P.

Итоговый отчёт строится по принципу согласованности источников. Чем больше баз подтверждают связь независимо, тем выше уровень уверенности: пять из пяти дают максимальный балл, а единичные упоминания помечаются как слабые.

Для полного прогона DisGeNET и OMIM требуют ключи в переменных DISGENET_API_KEY и OMIM_API_KEY. Если ключей нет, скрипты переключаются на открытые вызовы через Monarch и OpenTargets.

Какие задачи решает

  • Найти все заболевания, ассоциированные с конкретным геном
  • Определить гены-кандидаты по названию или коду болезни
  • Сравнить доказательства менделевских и комплексных связей по разным каталогам
  • Рассчитать согласованность данных между DisGeNET, OMIM и OpenTargets

Примеры запросов агенту

  • Проверь по базам ассоциации гена BRCA1 с заболеваниями и оцени согласованность
  • Собери список генов, связанных с синдромом Марфана, через Monarch и OMIM
  • Построй отчёт по связи гена CFTR с муковисцидозом с указанием баллов доказательности

Как установить tooluniverse-gene-disease-association

1Через Skills CLI (Vercel)

npx skills add mims-harvard/ToolUniverse --skill tooluniverse-gene-disease-association -a claude-code

2Вручную: положить файлы

~/.claude/skills/tooluniverse-gene-disease-association/SKILL.md

Папка скилла целиком; для одного проекта — .claude/skills/tooluniverse-gene-disease-association/

По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов

Вопросы о tooluniverse-gene-disease-association

Как установить tooluniverse-gene-disease-association в Claude Code?

Выполните команду npx skills add mims-harvard/ToolUniverse --skill tooluniverse-gene-disease-association -a claude-code в терминале.

Что делает этот скилл?

Он запрашивает базы данных генетики, сопоставляет идентификаторы и выводит сводный отчёт с числовой оценкой доказательств.

Обязательно ли получать API-ключи для работы?

Ключи нужны только для DisGeNET и OMIM. Без них скилл пропускает закрытые эндпоинты и берёт данные из открытых источников: OpenTargets, GenCC и Monarch.

Исходный текст

SKILL.md · Apache-2.0
---
name: tooluniverse-gene-disease-association
description: Gene-disease association analysis across DisGeNET, OpenTargets, Monarch, OMIM, GenCC, Orphanet, Gene2Phenotype (G2P). Cross-references multiple sources for evidence-graded association reports with concordance scoring (5/5 sources agree → strong, 1/5 → weak). Use for 'which diseases is gene X associated with' or 'which genes cause disease Y' queries with quantitative confidence, including clinical-panel-based curation (developmental disorders, cancer, cardiac, skeletal, eye, skin panels).
disable-model-invocation: true
---

# Gene-Disease Association Analysis

Systematically query and compare gene-disease associations across 6+ databases to produce a unified, evidence-graded report. Cross-references DisGeNET scores, OpenTargets evidence, Monarch Initiative cross-species data, OMIM Mendelian mappings, GenCC curated validity, and Orphanet rare disease links.

**IMPORTANT**: Always use English gene names and disease terms in tool calls. Respond in the user's language.

---

## LOOK UP, DON'T GUESS
When uncertain about any scientific fact, SEARCH databases first (PubMed, UniProt, ChEMBL, ClinVar, etc.) rather than reasoning from memory. A database-verified answer is always more reliable than a guess.

---

## Core Principles

1. **Report-first approach** - Create report file FIRST, then populate progressively
2. **Multi-database triangulation** - Query 4+ sources minimum, cross-validate
3. **Quantitative scoring** - Report numeric scores from each database
4. **Concordance analysis** - Count how many databases support each association and reason about independence
5. **Evidence reasoning** - Assess each association using evidence hierarchy, concordance, and mechanism plausibility
6. **Mendelian vs complex** - Distinguish monogenic (OMIM/Orphanet) from complex (GWAS/DisGeNET) associations
7. **Negative results documented** - "No association found in [database]" is informative

---

## Workflow Overview

```
Phase 1: Gene/Disease Identification & ID Resolution
  Resolve gene symbol to Ensembl ID, HGNC CURIE, MIM number
  OR resolve disease name to UMLS CUI, EFO ID, MONDO ID, ORPHA code
      |
Phase 2: DisGeNET Associations (scored, multi-evidence)
  Gene-disease association scores with evidence type filtering
      |
Phase 3: OpenTargets Associations (integrated evidence)
  Disease phenotypes and genetic associations from OpenTargets
      |
Phase 4: Monarch Initiative (cross-species evidence)
  Gene-disease associations integrating OMIM, ClinVar, model organisms
      |
Phase 5: Mendelian Disease Evidence (curated)
  OMIM gene-disease map, GenCC validity classifications, Orphanet rare diseases,
  Gene2Phenotype (G2P) clinical-panel assertions
      |
Phase 6: Variant-Disease Associations (optional, if gene query)
  DisGeNET variant-disease links, ClinVar pathogenic variants
      |
Phase 7: Evidence Synthesis
  Unified table, concordance scoring, confidence levels, final report
```

---

## Phase 1: Gene/Disease Identification & ID Resolution

```python
from tooluniverse import ToolUniverse
tu = ToolUniverse()
tu.load_tools()

# Gene query: resolve IDs
gene_info = tu.tools.MyGene_query_genes(query=f"symbol:{gene_symbol}", species="human",
    fields="symbol,ensembl.gene,entrezgene,name", size=5)  # -> ensembl_id
Monarch_search = tu.tools.MonarchV3_search(query=gene_symbol, category="biolink:Gene", limit=5)  # -> HGNC CURIE
omim_result = tu.tools.OMIM_search(query=gene_symbol, limit=5)  # -> MIM number
gene_summary = tu.tools.Harmonizome_get_gene(gene_symbol=gene_symbol)

# Disease query: resolve IDs
monarch_disease = tu.tools.MonarchV3_search(query=disease_name, category="biolink:Disease", limit=5)  # -> MONDO CURIE
mappings = tu.tools.MonarchV3_get_mappings(entity_id=mondo_id, limit=20)  # -> OMIM, ICD10, SNOMED, Orphanet
```

---

## Phase 2: DisGeNET Associations

> **API KEY REQUIRED**: DisGeNET tools require `DISGENET_API_KEY` environment variable. Without it, all DisGeNET calls will fail. Register at https://www.disgenet.org/api/#/Authorization for a free academic key.
> **Fallback if no key**: Skip this phase and rely on OpenTargets (Phase 3) + Monarch (Phase 4) which are free and cover much of the same data.

```python
# Gene -> diseases
disgenet_diseases = tu.tools.DisGeNET_search_gene(gene=gene_symbol, limit=20)
disgenet_gda = tu.tools.DisGeNET_get_gda(gene=gene_symbol, source="CURATED", min_score=0.3, limit=25)

# Disease -> genes (accepts name or UMLS CUI like "C0006142")
disgenet_genes = tu.tools.DisGeNET_search_disease(disease=disease_name, limit=20)
disgenet_ranked = tu.tools.DisGeNET_get_disease_genes(disease=disease_name, min_score=0.3, limit=50)
```

**Interpreting DisGeNET scores**: Higher scores reflect more evidence sources and stronger curation. Rather than memorizing cutoffs, ask: is this score driven by curated sources or text-mining? Use `source="CURATED"` to distinguish.

---

## Phase 3: OpenTargets Associations

```python
ot_diseases = tu.tools.OpenTargets_get_diseases_phenotypes_by_target_ensembl(ensemblId=ensembl_id)
ot_evidence = tu.tools.OpenTargets_target_disease_evidence(ensemblId=ensembl_id, efoId=efo_id)
# Both require pre-resolved Ensembl/EFO IDs. Use OpenTargets_multi_entity_search_by_query_string to discover IDs.
```

---

## Phase 4: Monarch Initiative Associations

```python
# Gene -> diseases (integrates OMIM, ClinVar, Orphanet, model organisms)
monarch_diseases = tu.tools.MonarchV3_get_associations(
    subject=hgnc_curie, category="biolink:CausalGeneToDiseaseAssociation", limit=20)
# Disease -> genes
monarch_genes = tu.tools.MonarchV3_get_associations(
    subject=mondo_id, category="biolink:CorrelatedGeneToDiseaseAssociation", limit=20)
histopheno = tu.tools.MonarchV3_get_histopheno(entity_id=mondo_id)  # phenotypes by body system
entity = tu.tools.MonarchV3_get_entity(entity_id=hgnc_curie)  # details, synonyms, xrefs
```

Note: `EnsemblPheno_get_by_gene` is another multi-source gene->phenotype
aggregator (GWAS catalog/ClinVar/OMIM/Orphanet), but is currently
returning live HTTP 500 errors upstream at Ensembl — verified, not a
ToolUniverse bug. Its variant-level counterpart, `EnsemblPheno_get_by_variant`,
does work; see `tooluniverse-variant-analysis`'s "Ensembl Phenotype
Associations" section. Prefer Monarch/G2P above until Ensembl's gene/term/
region endpoints recover.

---

## Phase 5: Mendelian Disease Evidence

> **API KEY REQUIRED**: OMIM tools require `OMIM_API_KEY`. Register at https://omim.org/api for academic access.
> **Fallback if no key**: Use Monarch Initiative (`biolink:CausalGeneToDiseaseAssociation` from Phase 4) which includes OMIM data without requiring a key. Also use GenCC (below) which is fully open.

```python
# OMIM: Mendelian gene-disease mapping (use gene MIM number, not phenotype MIM)
omim_entry = tu.tools.OMIM_get_entry(mim_number=mim_number)
omim_gene_map = tu.tools.OMIM_get_gene_map(mim_number=mim_number)
omim_clinical = tu.tools.OMIM_get_clinical_synopsis(mim_number=phenotype_mim)

# GenCC: curated validity (Definitive/Strong/Moderate/Limited/Disputed/Refuted)
gencc_result = tu.tools.GenCC_search_gene(gene_symbol=gene_symbol)  # handles gene renames
gencc_disease = tu.tools.GenCC_search_disease(disease="Marfan syndrome")  # word-tokenized matching
gencc_classifications = tu.tools.GenCC_get_classifications(gene_symbol="BRCA1", disease="breast cancer")

# Orphanet: rare disease associations (filter results by exact gene.symbol match)
orphanet_result = tu.tools.Orphanet_get_gene_diseases(gene_name=gene_symbol)

# Gene2Phenotype (G2P): EBI-curated gene-disease pairs organized into clinical panels
# (DD=developmental disorders, Cancer, Skeletal, Eye, Cardiac, Ear, Skin, Prenatal).
# Distinct from GenCC/OMIM: G2P assertions carry genotype (monoallelic/biallelic),
# molecular mechanism (loss/gain of function), and are grouped by clinical specialty panel
# rather than by disease name alone -- useful when the query maps naturally onto a panel
# (e.g. "what developmental-disorder genes are definitive-confidence" -> DD panel).
g2p_hits = tu.tools.G2P_search(query=gene_symbol)  # -> list of stable_ids, one per gene-disease pair
g2p_record = tu.tools.G2P_get_record(stable_id=g2p_hits["data"]["results"][0]["stable_id"])
# g2p_record includes genotype, molecular_mechanism, disease.ontology_terms (MONDO + OMIM
# cross-refs), confidence, and supporting publications (PMIDs) -- richer per-pair evidence
# than G2P_search's summary row.

# To enumerate every gene in a clinical panel (e.g. building a candidate gene list for a
# developmental-disorder patient), download the full panel rather than searching gene-by-gene:
g2p_panel_genes = tu.tools.G2P_download_panel(panel="DD")  # every curated DD gene-disease pair
g2p_panel_stats = tu.tools.G2P_get_panel(panel="dd")  # just counts (total_records, total_genes, by_confidence)
```

**G2P confidence tiers** (definitive > strong > moderate > limited) parallel GenCC's
Definitive/Strong/Moderate/Limited/Disputed/Refuted scale -- when both sources classify the
same gene-disease pair, concordant "definitive"/"Definitive" ratings from two independently
curated resources is stronger evidence than either alone. When they disagree, check whether
G2P's panel assignment matches the disease context: a gene rated "definitive" for the Cancer
panel says nothing about its status for, say, a cardiac phenotype.

---

## Phase 6: Variant-Disease Associations (Optional)

Run when the query is gene-based and variant-level evidence adds value.

```python
vda_result = tu.tools.DisGeNET_get_vda(gene=gene_symbol, limit=25)  # variant-disease links
clinvar_result = tu.tools.ClinVar_search_variants(gene=gene_symbol, max_results=20)
clinvar_detail = tu.tools.ClinVar_get_variant_details(variant_id="12345")  # detailed variant info
```

---

## Phase 7: Evidence Synthesis

### Unified Association Table

Compile all results into a single table per gene-disease pair:

```markdown
## Gene-Disease Associations for BRCA1

| Disease | DisGeNET Score | OpenTargets Score | Monarch | OMIM | GenCC | Orphanet | Sources |
|---------|---------------|-------------------|---------|------|-------|----------|---------|
| Breast cancer | 0.82 | 0.95 | Yes | #114480 | Definitive | ORPHA:227535 | 6/6 |
| Ovarian cancer | 0.78 | 0.91 | Yes | #604370 | Definitive | ORPHA:213500 | 6/6 |
| Pancreatic cancer | 0.35 | 0.42 | Yes | - | Moderate | - | 3/6 |
| Fanconi anemia | 0.45 | 0.38 | Yes | #605724 | Strong | ORPHA:84 | 5/6 |
```

### Reasoning Strategies for Evidence Evaluation

**Evidence strength reasoning**: A gene-disease association supported by multiple independent lines of evidence (genetic, functional, model organism) is stronger than one supported by a single study. Ask: how many independent sources support this link? Do they converge on the same mechanism?

**Genetic evidence hierarchy**: Mendelian segregation (gene mutation causes disease in family) > GWAS (statistical association in population) > candidate gene study (hypothesis-driven). The first proves causation. The second shows correlation. The third is hypothesis. OMIM/GenCC "Definitive" entries represent the top of this hierarchy; DisGeNET text-mining hits represent the bottom.

**Cross-database concordance**: If DisGeNET, OpenTargets, AND OMIM all link gene X to disease Y, that's strong concordance. If only one database shows the link, check why -- is it a single study indexed by that database? Concordance across databases does not equal independent evidence if they all cite the same primary study. Count the number of databases supporting each association, but reason about whether they represent truly independent evidence.

**Mechanism reasoning**: Knowing the gene's function helps evaluate the association. A gene encoding a liver enzyme being linked to liver disease is mechanistically plausible. The same gene being linked to a psychiatric disorder needs stronger evidence because the mechanism is less obvious. Use Harmonizome gene summaries and Monarch phenotype profiles to assess mechanistic plausibility.

---

## Common Patterns

- **Gene-centric**: MyGene ID resolution -> DisGeNET/OpenTargets/Monarch/OMIM/GenCC/Orphanet -> unified table ranked by concordance
- **Disease-centric**: MonarchV3 disease search -> DisGeNET disease genes -> Monarch/OMIM -> unified gene table ranked by evidence
- **Gene-disease pair**: Resolve both IDs -> query all databases for the specific pair -> deep evidence summary with variant-level data
- **Mendelian discovery**: OMIM_search -> OMIM_get_gene_map -> GenCC per gene -> Orphanet -> filter by curated validity
- **Cross-species**: Monarch associations (HP/MP/ZP phenotypes) + DisGeNET ANIMAL_MODELS source -> model organism evidence
- **Gene renames**: GenCC handles renames automatically via `_gene_matches()`. Other tools require current HGNC symbol from MyGene_query_genes.

---

## Troubleshooting

- **DisGeNET returns empty**: Check DISGENET_API_KEY. Try UMLS CUI instead of disease name.
- **OMIM returns no gene map**: Use the gene MIM number, not the phenotype MIM number.
- **Monarch returns no associations**: Verify CURIE format (HGNC:1100 not HGNC:BRCA1). Use MonarchV3_search first.
- **OpenTargets returns no diseases**: Verify Ensembl ID via MyGene_query_genes with `fields="ensembl.gene"`.
- **GenCC returns empty**: Not all genes have classifications. Check for gene renames.
- **GenCC disease search misses**: Simplify query (e.g., "breast cancer" not "hereditary breast and ovarian cancer syndrome").
- **Gene rename misses in non-GenCC tools**: Only GenCC handles renames automatically. Use MyGene_query_genes to confirm the current canonical symbol.
- **G2P_search returns 0 results for a real disease gene**: G2P only covers genes with at least one clinically curated assertion in one of its panels (DD, Cancer, Skeletal, Eye, Cardiac, Ear, Skin, Prenatal) -- it is not a genome-wide resource like DisGeNET. A miss here does not mean no association exists elsewhere; fall back to the other Phase 5 sources.
- **G2P panel name mismatch**: Panel names are case-insensitive short codes (`dd`, `cancer`, `cardiac`, `skeletal`, `eye`, `ear`, `skin`, `prenatal`), not full disease-area sentences -- check `G2P_get_panel` first if unsure of the exact code.

---

## Resources

For comprehensive disease reports: [tooluniverse-disease-research](../tooluniverse-disease-research/SKILL.md)
For rare disease diagnosis: [tooluniverse-rare-disease-diagnosis](../tooluniverse-rare-disease-diagnosis/SKILL.md)
For variant interpretation: [tooluniverse-variant-interpretation](../tooluniverse-variant-interpretation/SKILL.md)
For drug-target validation: [tooluniverse-drug-target-validation](../tooluniverse-drug-target-validation/SKILL.md)

© mims-harvard · Apache-2.0 · mims-harvard/ToolUniverse

Похожие скиллы

Все скиллы по теме «Наука и исследования» →
locus-to-gene-mapper-skill официальный OpenAI Скилл locus-to-gene-mapper-skill сопоставляет локусы GWAS с генами-кандидатами через цепочку биоинформатических баз данных. Помогает приоритизировать гены по названию фенотипа или списку генетических вариантов. CX Codex GP ChatGPT +1 13установок biological-expert personamanagmentlayer Скилл biological-expert настраивает агента на задачи биоинформатики, генетики и вычислительной биологии. Помогает анализировать последовательности ДНК и аннотировать мутации. CX Codex +1 235установок gnomad-database google-deepmind Скилл gnomad-database для Claude Code запрашивает частоты аллелей вариантов, метрики ограничений генов и структурные варианты из базы gnomAD через Python-скрипты с автоматическим соблюдением лимитов API. CX Codex +1 2 300установок biobankjapan-phewas-skill официальный OpenAI Скилл biobankjapan-phewas-skill запрашивает ассоциации BioBank Japan для одного генетического варианта по rsID, GRCh37 или GRCh38. Помогает быстро получить сводку фенотипов прямо в диалоге с моделью. CX Codex GP ChatGPT +1 13установок Open Targets opentargets Удалённый MCP-сервер к платформе Open Targets: поиск целей, болезней и препаратов плюс GraphQL-запросы к связям генов с заболеваниями. CX Codex +1 440установок scientific-thinking-biology Agents365-ai Скилл scientific-thinking-biology для Claude Code структурирует анализ биологических исследований и генетических данных. Помогает отличить ассоциацию от причинности, маркер от драйвера. CX Codex +1 296установок

Из того же набора136

Весь набор →
tooluniverse-epigenomics-chromatin Скилл tooluniverse-epigenomics-chromatin анализирует гистоновые модификации, доступность хроматина и связывание TF по данным ENCODE и других источников. CX Codex +1 241установок tooluniverse-plant-genomics Скилл tooluniverse-plant-genomics ищет растительные гены, метаболические пути и ортологи через PlantReactome, KEGG и Ensembl Plants. Помогает исследовать полиплоидные культуры и вторичный метаболизм растений. CX Codex +1 240установок tooluniverse-structural-proteomics Скилл tooluniverse-structural-proteomics для Claude Code собирает структурные и протеомные данные для проверки лекарственной мишени. Помогает оценить участки связывания, лиганды и уверенность структурных моделей. CX Codex +1 240установок tooluniverse-small-molecule-discovery Скилл tooluniverse-small-molecule-discovery для Claude Code ищет и анализирует малые молекулы: от названия соединения до структуры, активности, ADMET-свойств и доступности. CX Codex +1 240установок tooluniverse-protein-modification-analysis Скилл tooluniverse-protein-modification-analysis для Claude Code анализирует посттрансляционные модификации белков. Он подключается для поиска сайтов модификации, ферментов и связей с заболеваниями. CX Codex +1 239установок tooluniverse-pathway-disease-genetics Скилл tooluniverse-pathway-disease-genetics для Claude Code связывает GWAS-варианты с генами, биологическими путями и лекарственными мишенями. Ищет механизмы заболеваний и репозиционирование препаратов. CX Codex +1 239установок tooluniverse-lipidomics Скилл tooluniverse-lipidomics для Claude Code идентифицирует липиды, классифицирует их по LIPID MAPS и связывает с метаболическими путями и болезнями. Анализирует липидомику и ищет биомаркеры липидных нарушений. CX Codex +1 238установок tooluniverse-comparative-genomics Скилл tooluniverse-comparative-genomics для Claude Code ищет ортологов и оценивает консервативность генов при сравнении организмов разных видов. CX Codex +1 238установок

Спонсоры проекта

Компании, которые помогают держать аналитику открытой