tooluniverse-chemical-safety

What it does

This skill performs comprehensive chemical safety and toxicology assessments by combining AI-predicted toxicity endpoints, toxicogenomics data, regulatory safety information, and chemical-protein interaction networks to generate a structured safety report with evidence grading.

How it works

1. Phase 0: Compound Disambiguation – resolves the input to a canonical identity (name, SMILES, CID, ChEMBL ID) and collects basic properties (formula, weight, etc.).
2. Phase 1: Predictive Toxicology (ADMET-AI) – runs toxicity endpoints via ADMETAI_predict_toxicity, ADMETAI_predict_stress_response, and ADMETAI_predict_nuclear_receptor_activity for SMILES input; interprets results as Active/Inactive and grades them with [T3]. Highlights high-risk signals (e.g., AMES, DILI, hERG).
3. Phase 2: ADMET Properties – (if SMILES available) executes six tools to characterize Absorption, Distribution, Metabolism, Excretion, and physicochemical properties; assesses Lipinski violations and DDI risks.
4. Phase 3: Toxicogenomics (CTD) – queries CTD for chemical-gene interactions and chemical-disease associations; reports top interactions with evidence tiers [T2] or [T3] and highlights affected pathways.
5. Phase 4: Regulatory Safety (FDA Labels) – retrieves FDA label data (boxed warnings, contraindications, adverse reactions, warnings, nonclinical toxicology, carcinogenic/mutagenic data); marks all FDA findings as [T1].
6. Synthesis – aggregates evidence across phases to provide a final risk classification and data gaps, with explicit source attribution for each safety claim.

When to use it

Use when asked about any of the following: chemical toxicity, drug safety profiling, ADMET properties, environmental health risks, chemical hazard assessment, or toxicogenomic analysis. Triggers include asking for toxicity endpoints, safety profiles, ADMET properties, gene interactions, disease associations, regulatory safety, or environmental toxicology related to a chemical.

What it can touch

The skill utilizes tools for compound identity resolution, ADMET toxicity predictions, CTD queries, FDA label extraction, DrugBank and STITCH-related data, and regulatory data sources. It requires the claude-code tool.

Caveats

All toxicity predictions are computational [T3]. FDA data, when present, is treated as [T1] regulatory evidence. When evidence is ambiguous, risk is noted as requiring further investigation; explicit data gaps are reported as such. If a tool fails, the report notes prediction unavailability rather than failing the entire entry.