tooluniverse-noncoding-rna — скилл для Claude Code
Скилл tooluniverse-noncoding-rna помогает анализировать некодирующие РНК, искать мишени микроРНК, аннотировать lncRNA и проверять связи с болезнями через научные базы данных.
Apache-2.0 225 установок ★ 1 798 обновлён 2026-10-24
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-noncoding-rna -a claude-codeЧто делает tooluniverse-noncoding-rna
Скилл маршрутизирует биологические запросы по базам данных в зависимости от класса молекулы: miRBase для микроРНК, LNCipedia для длинных некодирующих РНК и RNAcentral для универсального поиска по URS-идентификаторам. В цепочку входят вызовы PubMed для верификации публикаций, Rfam для семейств РНК, DisGeNET для ассоциаций с заболеваниями и GTEx для проверки экспрессии в тканях.
Логика работы опирается на строгую градуировку доказательств: экспериментально подтверждённые мишени из репортерных тестов и CLIP-seq имеют приоритет перед алгоритмическими предсказаниями. Для circRNA скилл требует проверять гипотезы о функции губки микроРНК через литературу, а не принимать их на веру.
Какие задачи решает
- найти экспериментально подтверждённые мишени микроРНК
- аннотировать последовательность lncRNA и варианты транскриптов
- получить геномные координаты по идентификатору RNAcentral
- проверить тканеспецифичность экспрессии гена некодирующей РНК
Примеры запросов агенту
- Найди подтверждённые мишени для miR-21 через miRBase и литературу
- Какие lncRNA ассоциированы с раком молочной железы по DisGeNET?
- Получи геномные координаты и последовательность для RNAcentral URS00003B7674
Как установить tooluniverse-noncoding-rna
1Через Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-noncoding-rna -a claude-code2Вручную: положить файлы
~/.claude/skills/tooluniverse-noncoding-rna/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/tooluniverse-noncoding-rna/
1Через Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-noncoding-rna -a codex2Вручную: положить файлы
.agents/skills/tooluniverse-noncoding-rna/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-noncoding-rna -a cursor2Вручную: положить файлы
.cursor/skills/tooluniverse-noncoding-rna/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install mims-harvard/ToolUniverse skills/tooluniverse-noncoding-rna2Вручную: положить файлы
.github/skills/tooluniverse-noncoding-rna/SKILL.md1Через Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-noncoding-rna -a gemini-cli2Командой платформы
gemini skills install https://github.com/mims-harvard/ToolUniverse --path skills/tooluniverse-noncoding-rnaВручную: положить файлы
.devin/skills/tooluniverse-noncoding-rna/SKILL.mdСтарый путь .windsurf/skills/ тоже читается
Через Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-noncoding-rna -a opencodeClaude.ai
Упакуйте папку в zip и загрузите в Settings → Capabilities → Skills
По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов
Вопросы о tooluniverse-noncoding-rna
Как установить tooluniverse-noncoding-rna в Claude Code?
Выполните команду npx skills add mims-harvard/ToolUniverse --skill tooluniverse-noncoding-rna -a claude-code. Конфигурация запишется по пути ~/.claude/skills/tooluniverse-noncoding-rna/SKILL.md.
Что делает tooluniverse-noncoding-rna?
Организует пайплайн поиска и аннотации некодирующих РНК, обращаясь к внешним базам miRBase, LNCipedia, RNAcentral, MODOMICS и GTEx.
Подходит ли этот скилл для дифференциальной экспрессии mRNA?
Нет, для дифференциального анализа экспрессии мРНК предназначен отдельный модуль tooluniverse-rnaseq-deseq2.
Исходный текст
SKILL.md · Apache-2.0
---
name: tooluniverse-noncoding-rna
description: Non-coding RNA analysis — miRNAs (miRBase, miRDB targets), lncRNAs (LNCipedia, RNAcentral), circRNAs, snoRNAs, and other ncRNA classes. Distinct mechanisms per class — miRNAs repress mRNA; lncRNAs scaffold/decoy/enhance. Use for ncRNA function prediction, miRNA-target prediction, lncRNA functional annotation, and ncRNA-disease association queries.
disable-model-invocation: true
---
# Non-Coding RNA Analysis
Pipeline for identifying, annotating, and interpreting non-coding RNAs and their biological roles. Covers microRNAs (miRNAs), long non-coding RNAs (lncRNAs), and other ncRNA classes.
**Key principles**:
1. **Class determines function** — miRNAs repress mRNA translation; lncRNAs have diverse mechanisms (scaffolds, guides, decoys, enhancers); rRNAs/tRNAs are structural
2. **Targets matter more than the ncRNA itself** — for miRNAs, the regulated mRNA targets determine the phenotype
3. **Expression context is critical** — ncRNAs are highly tissue/cell-type specific
4. **Conservation indicates function** — deeply conserved ncRNAs (miR-let-7, MALAT1) have well-established roles
5. **Evidence grading** — T1: validated targets (reporter assay, CLIP-seq), T2: high-confidence computational prediction, T3: expression correlation, T4: sequence-based prediction only
**Type-based reasoning — look up, don't guess**:
Non-coding RNA function depends on type: miRNA silences target mRNAs (look up targets in miRTarBase/TargetScan), lncRNA has diverse functions (scaffolding, guiding, decoying — check literature for the specific lncRNA), circRNA may sponge miRNAs.
For any ncRNA query: first identify the class from the name/sequence, then select the appropriate evidence source. Do not assume function based on name alone — a gene named "LINC" may have a characterized mechanism, or none at all. Always search PubMed for the specific ncRNA before interpreting. For miRNAs, validated targets (T1) from miRTarBase outweigh any computational prediction — a predicted target with no experimental support is a hypothesis, not a finding. For lncRNAs, mechanism is almost always determined by experimental studies; use `PubMed_search_articles` with the lncRNA name + "mechanism" or "function" to find relevant evidence. For circRNAs, miRNA sponging is the most common proposed mechanism but is frequently over-claimed — look for CLIP-seq or reporter assay evidence before asserting it.
---
## When to Use
- "What are the targets of miR-21?"
- "Find lncRNAs associated with breast cancer"
- "Is this lncRNA conserved across species?"
- "What miRNAs regulate TP53?"
- "Annotate these non-coding RNA IDs"
- "Which miRNAs are biomarkers for [disease]?"
**Not this skill**: For mRNA expression analysis, use `tooluniverse-rnaseq-deseq2`. For CRISPR screens, use `tooluniverse-crispr-screen-analysis`.
---
## Core Tools
| Tool | Use For |
|------|---------|
| `miRBase_search_mirna` | Search miRNAs by name, accession, or sequence |
| `miRBase_get_mirna` | Detailed miRNA info (sequence, genomic location, family) |
| `miRBase_get_mirna` | Mature miRNA sequences and annotations |
| `PubMed_search_articles` | Search for validated miRNA targets in literature (e.g., "miR-21 target validation") |
| `LNCipedia_search_lncrna` | Search lncRNAs by name, gene symbol, or transcript ID |
| `LNCipedia_get_lncrna` | Detailed lncRNA transcript info (sequence, structure, conservation) |
| `LNCipedia_get_lncrna_xrefs` | lncRNA gene info with all transcript variants |
| `LNCipedia_search_ncrna_by_type` | List all transcripts for a lncRNA gene |
| `LNCipedia_get_lncrna_publications` | lncRNA sequence (FASTA format) |
| `RNAcentral_search` | Search all ncRNA types across databases |
| `RNAcentral_get_by_accession` | Detailed ncRNA annotations from 40+ databases |
| `RNAcentral_get_genome_locations` | Genomic coordinates (chrom/strand/start/end, assembly) of a URS accession in a given organism — different question from `get_by_accession` (annotation, not coordinates) |
| `RNAcentral_get_sequence` | Raw FASTA nucleotide sequence for a URS accession |
| `MODOMICS_search_modifications` / `MODOMICS_list_modifications` | Look up RNA chemical modifications (m6A, pseudouridine, m5C, etc.) by name/formula, or browse all 430+ |
| `Rfam_get_family` | RNA family details (structure, alignment, species distribution) |
| `Rfam_search_sequence` | Search RNA families by keyword |
| `DisGeNET_search_gene` | ncRNA-disease associations |
| `PubMed_search_articles` | ncRNA literature |
| `GTEx_get_median_gene_expression` | Tissue expression of ncRNA genes |
---
## Workflow
```
Phase 0: ncRNA Identity & Classification
Name/ID → miRBase/LNCipedia/RNAcentral → class, sequence, genomic location
|
Phase 1: Target & Interaction Analysis
miRNA → target mRNAs; lncRNA → interacting proteins/RNAs/chromatin
|
Phase 2: Expression & Tissue Specificity
GTEx/GEO → where is it expressed? Tissue-specific or ubiquitous?
|
Phase 3: Disease Associations
DisGeNET/PubMed/CTD → ncRNA-disease links with evidence
|
Phase 4: Functional Interpretation
Pathway enrichment of targets → biological role → clinical significance
```
### Phase 0: ncRNA Identity & Classification
ncRNA classes by size and database:
- **miRNA** (~22 nt, miRBase): Post-transcriptional silencing via 3'UTR binding
- **lncRNA** (>200 nt, LNCipedia): Diverse — chromatin remodeling, transcription regulation, miRNA sponges
- **rRNA** (120-5000 nt, RNAcentral/Rfam): Ribosome components
- **tRNA** (~76 nt, RNAcentral): Amino acid delivery
- **snoRNA** (60-300 nt, Rfam): rRNA modification (methylation, pseudouridylation)
- **snRNA** (~150 nt, Rfam): Spliceosome components
- **piRNA** (26-31 nt, RNAcentral): Transposon silencing in germline
- **circRNA** (variable, RNAcentral): miRNA sponges, protein scaffolds (experimental evidence required)
**Identification workflow**:
- Name starts with `miR-` or `hsa-mir-` → search miRBase
- Name starts with `LINC`, `MALAT`, `HOTAIR`, `XIST`, or ends in `-AS1` → search LNCipedia
- Any ncRNA type → search RNAcentral (aggregates all databases)
- RNA family question → search Rfam
**Genomic location and raw sequence** (once you have a URS accession from
`RNAcentral_search`): `RNAcentral_get_by_accession` gives cross-database
*annotation*; `RNAcentral_get_genome_locations`/`RNAcentral_get_sequence`
are the separate tools for genomic *coordinates* and the raw *FASTA
sequence* respectively — verified live:
```
tu.run_tool("RNAcentral_get_genome_locations", {"urs_id": "URS00003B7674", "taxid": 9606})
# -> locations[] with chromosome, strand, start, end, assembly_id (GRCh38), gca_accession
tu.run_tool("RNAcentral_get_sequence", {"urs_id": "URS00003B7674"})
# -> real verified response for let-7a-5p:
# {"fasta_header": "URS00003B7674 ncRNA from 61 species",
# "sequence": "UGAGGUAGUAGAUUGUAUAGUU", "length": 22}
```
**RNA chemical modifications (MODOMICS)** — for epitranscriptomic questions
(m6A, pseudouridine, m5C, etc.), especially relevant for snoRNA-guided rRNA
modification and mass-spec-based modification identification:
```
tu.run_tool("MODOMICS_search_modifications", {"query": "pseudouridine"})
# -> matches by name/short-name/formula, e.g. id 8 "1-methyl-3-(3-amino-
# 3-carboxypropyl)pseudouridine"
tu.run_tool("MODOMICS_list_modifications", {"limit": 20})
# -> browse all 430+ known modifications with formula/mass/reference_moiety
```
**`MODOMICS_get_modification` is broken — verified live, do not use for a
specific ID lookup.** Calling it with `modification_id` 58, 78, or 22 (all
distinct, well-known modifications per the tool's own description: m6A,
pseudouridine, m5C) returns the identical hardcoded record every time (id 2,
"5,2'-O-dimethylcytidine") — the parameter is silently ignored. `search`/
`list` correctly vary by query and return `name`/`short_name`/`formula`/
`mass_avg`, which covers most needs; only SMILES/monoisotopic-mass/
product-ion detail (unique to `get_modification`) is currently unobtainable
for a specific ID — say so rather than fabricating those fields.
### Phase 1: Target & Interaction Analysis
**For miRNAs** — the targets determine the biology:
**PRIMARY TOOL**: `ENCORI_get_miRNA_targets` looks up miRNA-target interactions from ENCORI/starBase (CLIP-seq-supported + computationally predicted), no download needed:
1. **miRNA → targets**: `ENCORI_get_miRNA_targets(mirna="hsa-miR-21-5p", clip_min=1)` — each hit reports `clip_experiments` (CLIP-seq support; higher = stronger experimental evidence) and `predicted_by` (which programs call it). Results are ranked by CLIP support, so the top rows are the best-supported targets.
2. **gene → miRNAs**: `ENCORI_get_miRNA_targets(gene="TP53")` — which miRNAs target a gene.
Supporting/fallback approaches:
3. **Literature** (for mechanism/validation context): `PubMed_search_articles(query="miR-21 target validation luciferase")`
4. **Cross-references**: `miRBase_get_mirna_xrefs(accession="MIMAT0000076")`
5. **For novel miRNAs** not in ENCORI: search PubMed for "[miRNA] target".
Well-studied miRNA targets (for common oncomiRs/tumor suppressors):
- **miR-21**: PTEN, PDCD4, TPM1, RECK, SPRY1, SPRY2, BTG2
- **miR-155**: SOCS1, SHIP1, AID, TP53INP1
- **miR-122**: SLC7A1, ADAM17 (also HCV IRES cofactor)
- **let-7**: RAS, HMGA2, MYC, LIN28
**Target interpretation framework**:
- **Validated** (T1): Luciferase reporter, CLIP-seq, degradome-seq — base conclusions on these
- **High-confidence prediction** (T2): TargetScan conserved sites, DIANA-microT score > 0.9 — support validated findings
- **Prediction only** (T3-T4): miRanda, PicTar, RNA22 — hypothesis generation only; do not report as findings
**For lncRNAs** — the mechanism varies:
| lncRNA Mechanism | Example | How to Investigate |
|---|---|---|
| **Chromatin modifier** | HOTAIR, XIST | Check interacting proteins (PRC2, LSD1) via PubMed |
| **Transcription regulator** | NEAT1, MEG3 | Check nearby genes (cis-regulation) via genomic location |
| **miRNA sponge** | MALAT1, circRNAs | Search for miRNA binding sites |
| **Scaffold** | NKILA, BCAR4 | Check protein interactions |
| **Enhancer RNA** | eRNAs | Check ENCODE enhancer annotations |
### Phase 2: Expression & Tissue Specificity
```python
GTEx_get_median_gene_expression(gene_symbol="MIR21") # miRNA host gene expression
# Note: GTEx measures RNA-seq; miRNA expression may need miRNA-seq data from GEO
```
**Interpretation**: Tissue-restricted ncRNAs are often functionally important in that tissue. Ubiquitous ncRNAs (like MALAT1) tend to have housekeeping roles.
### Phase 3: Disease Associations
```python
DisGeNET_search_gene(query="MIR21") # miR-21 disease associations
PubMed_search_articles(query="miR-21 biomarker cancer")
```
**Key ncRNA-disease associations** (well-established T1 examples — always verify via DisGeNET or PubMed for the specific ncRNA):
- miR-21: OncomiR in multiple cancers; targets PTEN, PDCD4, TPM1 (hundreds of T1 studies)
- miR-155: B-cell lymphoma, inflammation — immune regulation
- miR-122: Hepatitis C liver disease — HCV replication cofactor; therapeutic target (miravirsen)
- let-7 family: Lung cancer, stem cell differentiation — tumor suppressor targeting RAS, HMGA2
- HOTAIR: Breast/colorectal cancer — recruits PRC2, promotes metastasis
- MALAT1: Lung cancer/metastasis — splicing regulation
- XIST: X-inactivation, cancer — chromatin silencing
- H19: Beckwith-Wiedemann syndrome, cancer — imprinted lncRNA, miR-675 host
- ANRIL: CVD, diabetes, cancer — CDKN2A/B locus regulation (GWAS-validated)
### Phase 4: Functional Interpretation
After identifying miRNA targets (Phase 1), run pathway enrichment:
```python
# Collect validated target gene symbols
targets = ["PTEN", "PDCD4", "TPM1", "RECK", "SPRY1"] # miR-21 targets
# Pathway enrichment
ReactomeAnalysis_pathway_enrichment(identifiers="PTEN PDCD4 TPM1 RECK SPRY1")
STRING_get_network(identifiers="PTEN\rPDCD4\rTPM1\rRECK\rSPRY1", species=9606)
```
**Interpretation**: If miR-21 targets are enriched in apoptosis and PI3K-AKT signaling → miR-21 is an oncomiR that promotes survival by simultaneously suppressing multiple tumor suppressors.
**Report structure**:
1. **ncRNA Identity** — class, sequence, genomic location, conservation
2. **Targets/Interactions** — validated targets with evidence grades
3. **Expression Profile** — tissue specificity, disease-specific expression changes
4. **Disease Associations** — evidence-graded disease links
5. **Pathway Analysis** — enriched pathways among targets
6. **Mechanistic Model** — how this ncRNA contributes to disease biology
7. **Clinical Potential** — biomarker utility, therapeutic target potential (antagomirs, ASOs)
---
## Limitations
### Computational Procedure: TargetScan Predicted Targets (Download-and-Process)
TargetScan provides the best computational miRNA target predictions but has no REST API. Download and process locally:
```python
# Step 1: Download TargetScan predicted targets (one-time, ~10MB zipped)
# URL: https://www.targetscan.org/vert_80/vert_80_data_download/Summary_Counts.default_predictions.txt.zip
import pandas as pd
import zipfile, io, requests
url = "https://www.targetscan.org/vert_80/vert_80_data_download/Summary_Counts.default_predictions.txt.zip"
resp = requests.get(url, timeout=60)
with zipfile.ZipFile(io.BytesIO(resp.content)) as z:
fname = z.namelist()[0]
df = pd.read_csv(z.open(fname), sep='\t')
# Step 2: Query for a specific miRNA family
mirna = "miR-21-5p" # or "miR-21/590-5p" (TargetScan uses family names)
targets = df[df['miRNA Family'].str.contains("miR-21", case=False, na=False)]
# Step 3: Rank by cumulative weighted context++ score
targets_ranked = targets.sort_values('Cumulative weighted context++ score', ascending=True)
print(f"Top 20 predicted targets of {mirna}:")
for _, row in targets_ranked.head(20).iterrows():
print(f" {row['Target Gene']:10s} score={row['Cumulative weighted context++ score']:.3f} "
f"sites={row['Total num conserved sites']}")
```
**Interpretation**: More negative context++ score = stronger predicted repression. Conserved sites (>1) are higher confidence.
### Computational Procedure: miRTarBase Validated Targets (Download-and-Process)
miRTarBase has Cloudflare protection blocking programmatic access. Use the R/Bioconductor data package or bulk download:
```python
# Option 1: Download from miRTarBase bulk export (requires browser download first)
# Go to: https://mirtarbase.cuhk.edu.cn/~miRTarBase/miRTarBase_2025/
# Download: hsa_MTI.xlsx (human miRNA-target interactions)
# Option 2: Use the GitHub data dump
# https://github.com/jorainer/mirtarbase — R package with cached data
# Once you have the file:
import pandas as pd
mti = pd.read_excel("hsa_MTI.xlsx") # or read_csv if TSV
# Filter for your miRNA
mir21_targets = mti[mti['miRNA'].str.contains('hsa-miR-21', case=False, na=False)]
print(f"miR-21 validated targets: {len(mir21_targets)}")
# Filter by evidence strength
strong = mir21_targets[mir21_targets['Support Type'].str.contains(
'Luciferase|Reporter|Western|CLIP', case=False, na=False
)]
print(f" Strong evidence (reporter/CLIP): {len(strong)}")
for _, row in strong.head(10).iterrows():
print(f" {row['Target Gene']:10s} — {row['Support Type']}")
```
**When download is not available**: Use the built-in reference table in Phase 1 for well-studied miRNAs, or search PubMed for validated targets.
---
## Limitations
- **miRNA target prediction is noisy** — even the best algorithms have >50% false positive rates; always prioritize experimentally validated targets
- **lncRNA function is poorly characterized** — only ~5% of annotated lncRNAs have known functions
- **Expression measurement varies** — miRNA-seq, RNA-seq, and microarray capture different ncRNA classes; check the assay type
- **Species differences** — miRNAs are often conserved but lncRNAs are frequently species-specific; cross-species lncRNA comparisons are unreliable
© mims-harvard · Apache-2.0 · mims-harvard/ToolUniverse