tooluniverse-functional-genomics-screens — Claude Code skill
tooluniverse-functional-genomics-screens is a skill for ranking hits from genetic screens (CRISPR-KO, CRISPRi, shRNA) using essentiality, pathway, and clinical data.
Apache-2.0 225 installs ★ 1,798 updated 2026-10-24
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-functional-genomics-screens -a claude-codeWhat tooluniverse-functional-genomics-screens does
This skill runs a six-phase workflow to turn screen results into actionable drug targets. You provide a gene list, screen type, and cell model; the skill returns genes ranked by six criteria: selective essentiality, pathway convergence, druggability, clinical evidence, evolutionary constraint, and supporting literature.
Each gene is validated in DepMap (cell-line dependency scores), gnomAD (loss-of-function tolerance), and UniProt. The skill then searches Reactome and STRING for enriched pathways, queries DGIdb for existing drugs against each target, and gathers clinical evidence from CIViC and COSMIC. Missing DepMap per-line scores fall back to a local script from the bundled tooluniverse-cell-line-profiling skill.
The skill enforces a "look up, don't guess" principle: uncertain claims trigger database searches first. For large lists (>500 genes), pathway enrichment runs on the full set, then per-gene scoring focuses on the top 50. If gnomAD is overloaded, the skill retries or continues without those scores and flags the gap.
What it helps with
- Filter pan-essential genes and identify selective dependencies in a screen
- Rank hits by likelihood of being real drug targets
- Discover which genes co-enrich in the same pathway and understand functional context
- Collect cross-database evidence (DepMap, gnomAD, DGIdb, CIViC) for one gene
- Map screen hits to known clinical mutations and approved drugs
Example prompts for the agent
- Analyze CRISPR-KO hits from a lung-cancer screen: TP53, KRAS, EGFR, STK11
- Take 200 genes from an shRNA screen, find enriched pathways, and remove pan-essential hits
- For a CRISPRi screen in HeLa cells, rank the top 30 by selectivity and druggability
How to install tooluniverse-functional-genomics-screens
1Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-functional-genomics-screens -a claude-code2Manually: put the files in place
~/.claude/skills/tooluniverse-functional-genomics-screens/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/tooluniverse-functional-genomics-screens/
1Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-functional-genomics-screens -a codex2Manually: put the files in place
.agents/skills/tooluniverse-functional-genomics-screens/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-functional-genomics-screens -a cursor2Manually: put the files in place
.cursor/skills/tooluniverse-functional-genomics-screens/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install mims-harvard/ToolUniverse skills/tooluniverse-functional-genomics-screens2Manually: put the files in place
.github/skills/tooluniverse-functional-genomics-screens/SKILL.md1Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-functional-genomics-screens -a gemini-cli2Platform CLI
gemini skills install https://github.com/mims-harvard/ToolUniverse --path skills/tooluniverse-functional-genomics-screensManually: put the files in place
.devin/skills/tooluniverse-functional-genomics-screens/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add mims-harvard/ToolUniverse --skill tooluniverse-functional-genomics-screens -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about tooluniverse-functional-genomics-screens
How do I install tooluniverse-functional-genomics-screens in Claude Code?
Run npx skills add mims-harvard/ToolUniverse --skill tooluniverse-functional-genomics-screens -a claude-code in the terminal. The skill will be loaded at ~/.claude/skills/tooluniverse-functional-genomics-screens/SKILL.md.
What does tooluniverse-functional-genomics-screens do?
It takes a list of genes from a genetic screen and computes a score for each across six dimensions: selective essentiality, pathway enrichment, drug availability, clinical evidence, evolutionary constraint, and literature support. Output is a ranked hit list ready for validation or development.
Which databases does the skill integrate?
DepMap (per-line gene essentiality), gnomAD (loss-of-function constraint), Reactome and STRING (pathways and interactions), DGIdb (drugs and targets), CIViC and COSMIC (clinical mutations), and UniProt (gene function).
Source text
SKILL.md · Apache-2.0
--- name: tooluniverse-functional-genomics-screens description: Interpret hits from CRISPR-KO/CRISPRi/shRNA screens by integrating DepMap essentiality, gnomAD constraint scores, pathway context (Reactome, STRING), druggability (DGIdb), and clinical evidence (CIViC, COSMIC). Use for screen-hit prioritization, essentiality ranking, and turning a list of screen hits into a prioritized target shortlist. disable-model-invocation: true --- # Functional Genomics Screen Interpretation Pipeline for validating and prioritizing hits from genetic screens (CRISPR-KO, CRISPRi, shRNA) by integrating essentiality (DepMap), constraint (gnomAD), pathways (Reactome, STRING), druggability (DGIdb), and clinical evidence (CIViC, COSMIC). **Guiding principles**: 1. **Hits are hypotheses** -- screen results contain false positives; validate through orthogonal evidence 2. **Selectivity matters** -- pan-essential genes are poor drug targets; context-specific essentiality is high-value 3. **Pathway over gene** -- enriched pathways are more robust than individual hits 4. **Druggability is practical** -- prioritize chemically modulable targets 5. **English-first queries** -- use English gene names in tool calls ## LOOK UP, DON'T GUESS When uncertain about any scientific fact, SEARCH databases first. --- ## 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. ## Workflow ``` Phase 0: Input Processing → gene list, screen type, cell line, disease context Phase 1: Hit Validation → DepMap dependency, gnomAD constraint, UniProt function Phase 2: Pathway & Network → Reactome enrichment, STRING network, functional clusters Phase 3: Druggability → DGIdb interactions, druggable categories, PharmacoDB Phase 4: Clinical Evidence → CIViC, COSMIC mutations Phase 5: Literature → PubMed for key hits Phase 6: Prioritized Report → ranked target list with multi-dimensional scoring ``` --- ## Phase Details ### Phase 1: Hit Validation **Tools**: - `DepMap_get_gene_dependencies(gene_symbol=...)` -- returns gene metadata only (NOT per-cell-line scores) - `DepMap_search_cell_lines(query=...)` -- cell line metadata - `gnomad_get_gene_constraints(gene_symbol=...)` -- pLI, LOEUF (may return "Service overloaded") - `UniProt_get_function_by_accession(accession=...)` -- function summary **Classification**: Pan-essential (>90% lines), Selectively essential (specific lineages), Context-specific (screen model only). Chronos < -0.5 = likely essential, < -1.0 = strongly essential. **DepMap per-cell-line Chronos scores**: `DepMap_get_gene_dependencies` returns metadata only. For the actual per-cell-line scores, use the bundled script in the cell-line-profiling skill — `tooluniverse-cell-line-profiling/scripts/depmap_gene_dependency.py` (downloads the current DepMap Public CRISPRGeneEffect.csv once, cached; queries by `gene` or `cell-line`): ```bash python depmap_gene_dependency.py gene KRAS --lineage Lung --top 20 # most-dependent lines python depmap_gene_dependency.py cell-line A375 --top 25 # genes the line needs ``` Chronos < -0.5 ≈ dependency, < -1.0 strongly essential. Fallback if you can't run it: gnomAD constraint + `PubMed_search_articles(query="[gene] CRISPR screen [cancer]")`. ### Phase 2: Pathway & Network - `ReactomeAnalysis_pathway_enrichment(identifiers="TP53 BRCA1 EGFR")` -- space-separated string - `STRING_get_network(identifiers="GENE1\rGENE2\rGENE3", species=9606)` -- carriage-return separated - `STRING_functional_enrichment(identifiers=..., species=9606)` -- GO/KEGG enrichment ### Phase 3: Druggability - `DGIdb_get_drug_gene_interactions(genes=["EGFR","BRAF"])` -- drug-gene interactions - `DGIdb_get_gene_druggability(genes=[...])` -- categories (kinase, GPCR, etc.) - For high-priority hits, also search `search_clinical_trials` and PubMed for novel inhibitors not yet in DGIdb. ### Phase 4: Clinical Evidence - `civic_search_evidence_items(molecular_profile=gene)` -- NOT `query` - `COSMIC_get_mutations_by_gene(gene_name=...)` -- somatic mutation frequency ### Phase 6: Prioritized Report **Scoring (0-18)**: | Criterion | Score 3 | Score 0 | |-----------|---------|---------| | Selective essentiality | <-0.5 in disease AND >-0.2 elsewhere | >-0.2 (not essential) | | Pathway convergence | 3+ hits same pathway | Isolated hit | | Druggability | Approved drug exists | Not druggable | | Clinical evidence | CIViC therapeutic | No clinical data | | Constraint | pLI >0.9 | No data | | Literature | Multiple validation studies | No publications | **Tiers**: T1 (15-18) high-confidence, T2 (10-14) promising, T3 (5-9) speculative, T4 (<5) likely false positive. --- ## Edge Cases - **gnomAD overloaded**: Retry once, proceed without, note gap - **Gene not in DepMap**: Fall back to gnomAD + UniProt - **Large hit lists (>500)**: Pathway enrichment on full list; per-gene analysis on top 50 - **Non-cancer screens**: DepMap less informative; weight constraint/pathway more - **shRNA vs CRISPR**: Higher validation bar for shRNA (off-target effects) ## Limitations - DepMap is cancer-centric (~1000 cancer lines) - No raw screen analysis (use MAGeCK/BAGEL upstream) - STRING interactions are associations, not causal
© mims-harvard · Apache-2.0 · mims-harvard/ToolUniverse