Gene Liability Evaluation (ToolUniverse Claude Skill)

A ToolUniverse agent skill that answers “how dangerous is it to drug this target?” by scoring five independent safety dimensions into a single 0–100 liability figure and a recommended modulation strategy.


Type	Claude Skill (one of ToolUniverse’s pre-built agent skills)
Supplier	Zitnik Lab, Harvard Medical School
Availability	GA — part of the ToolUniverse skills collection (`skills/tooluniverse-gene-liability/`)
Pricing	Free / OSS (Apache-2.0); the underlying gnomAD, GTEx, Open Targets and DepMap queries are free public APIs
Capabilities	Read-only — drives ToolUniverse tool calls; no data writes

How to install

This skill calls ToolUniverse tools, so the ToolUniverse MCP server must be installed first (see the ToolUniverse page). Simplest registration:

claude mcp add --transport stdio tooluniverse -- uvx tooluniverse

Then add the skills:

Claude Code — install the whole skill collection (the skill resolves as tooluniverse-gene-liability):
```
npx skills add mims-harvard/ToolUniverse
```
Manual / other agents — copy just this skill directory into your skills folder:
```
git clone https://github.com/mims-harvard/ToolUniverse
cp -r ToolUniverse/skills/tooluniverse-gene-liability ~/.claude/skills/
```
(replace ~/.claude/skills/ with your agent’s skills directory if you are not using Claude Code/Desktop.)

What it does

Runs a five-phase target-safety assessment:

Gene resolution — MyGene_query_genes normalises whatever identifier you supplied (symbol, Ensembl, UniProt, Entrez) to a canonical gene.
Evidence gathering across five dimensions — human genetic constraint (gnomad_get_gene_constraints), mammalian knockout phenotype (OpenTargets_get_biological_mouse_models_by_ensemblID), critical-organ expression (GTEx_get_median_gene_expression), observed on-target clinical effects (OpenTargets_get_target_safety_profile_by_ensemblID), and cellular essentiality (DepMap_get_gene_dependencies). OpenTargets_get_associated_drugs_by_target_ensemblID supplies precedent from existing drugs against the target.
Dimension scoring — each dimension earns 0–25 points, graded by evidence tier T1–T4 so that direct human evidence outweighs inference.
Score calculation — liability is 100 × points earned / available weight, so dimensions with no data are excluded from the denominator rather than silently scored as safe.
Strategy translation — the score maps to a concrete recommendation: full inhibition, partial modulation, transient dosing, tissue-specific delivery, or deprioritise the target.

Constraint scoring is anchored on gnomAD: pLI ≥ 0.9 combined with LOEUF ≤ 0.35 takes the full 25 points for that dimension, and embryonic lethality in the mouse knockout takes the full 25 for phenotype.

Primary use cases: target triage before a discovery programme commits, on-target toxicity risk memos, deciding between a degrader and a partial inhibitor.

Notes

It is a reasoning layer over ToolUniverse; without the MCP server registered, the tool calls fail. The output is an on-target liability estimate — off-target chemical liabilities (hERG, AMES, DILI) are the domain of ADMET Prediction and ADMETlab MCP Server, and the skill does not attempt them.

Complements Drug Target Validation, which asks whether a target is efficacious, where this one asks whether it is safe. The individual evidence sources are catalogued separately if you want to query them directly: gnomAD, GTEx, Open Targets, DepMap, MyGene.info. ToolUniverse ships ~68 such skills; other workflows are catalogued separately.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.