Tooluniverse Chemical Safety

Chemical Safety & Toxicology Assessment

Toxicity assessment: identify the chemical, check known hazards (GHS, IARC), then look for ADMET predictions. Dose makes the poison — always consider exposure level, as a compound that is toxic at high doses may be safe at relevant exposures. Distinguish between acute toxicity (LD50, GHS category) and chronic hazards (carcinogenicity, endocrine disruption) — they require different risk management approaches. Computational predictions (ADMETAI) are T3 evidence and must be anchored by experimental data from PubChemTox or FDA labels wherever available. When evidence conflicts between prediction and experiment, always defer to the experimental finding.

LOOK UP DON'T GUESS: never assume GHS categories, IARC classification, or CTD disease links — always call PubChemTox and CTD tools to retrieve current classifications before reporting.

Comprehensive chemical safety analysis integrating predictive AI models, curated toxicogenomics databases, regulatory safety data, and chemical-biological interaction networks.

When to Use This Skill

Triggers:

• "Is this chemical toxic?" / "Assess the safety profile of [drug/chemical]"
• "What are the ADMET properties of [SMILES]?"
• "What genes does [chemical] interact with?" / "What diseases are linked to [chemical] exposure?"
• "Drug safety assessment" / "Environmental health risk" / "Chemical hazard profiling"

Use Cases: 1. Predictive Toxicology: AI-predicted endpoints (AMES, DILI, LD50, carcinogenicity, hERG) via SMILES 2. ADMET Profiling: Absorption, distribution, metabolism, excretion, toxicity 3. Toxicogenomics: Chemical-gene-disease mapping from CTD 4. Regulatory Safety: FDA label warnings, contraindications, adverse reactions 5. Drug Safety: DrugBank safety + FDA labels combined 6. Chemical-Protein Interactions: STITCH-based interaction networks 7. Environmental Toxicology: Chemical-disease associations for contaminants

---

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.

KEY PRINCIPLES

1. Report-first approach - Create report file FIRST, then populate progressively 2. Tool parameter verification - Verify params via get_tool_info before calling unfamiliar tools 3. Evidence grading - Grade all safety claims by evidence strength (T1-T4) 4. Citation requirements - Every toxicity finding must have inline source attribution 5. Mandatory completeness - All sections must exist with data or explicit "No data" notes 6. Disambiguation first - Resolve compound identity (name -> SMILES, CID, ChEMBL ID) before analysis 7. Negative results documented - "No toxicity signals found" is data; empty sections are failures 8. Conservative risk assessment - When evidence is ambiguous, flag as "requires further investigation" 9. English-first queries - Always use English chemical/drug names in tool calls

---

Evidence Grading System (MANDATORY)

Evidence grades MUST appear in: Executive Summary, Toxicity Predictions, Regulatory Safety, Chemical-Gene Interactions, Risk Assessment.

---

Core Strategy: 8 Research Phases

Chemical/Drug Query
|
+-- PHASE 0: Compound Disambiguation (ALWAYS FIRST)
|   Resolve name -> SMILES, PubChem CID, ChEMBL ID, formula, weight
|
+-- PHASE 1: Predictive Toxicology (ADMET-AI)
|   AMES, DILI, ClinTox, carcinogenicity, LD50, hERG, skin reaction
|   Stress response pathways, nuclear receptor activity
|
+-- PHASE 2: ADMET Properties
|   BBB penetrance, bioavailability, clearance, CYP interactions, physicochemical
|
+-- PHASE 3: Toxicogenomics (CTD)
|   Chemical-gene interactions, chemical-disease associations
|
+-- PHASE 4: Regulatory Safety (FDA Labels)
|   Boxed warnings, contraindications, adverse reactions, nonclinical tox
|
+-- PHASE 5: Drug Safety Profile (DrugBank)
|   Toxicity data, contraindications, drug interactions
|
+-- PHASE 6: Chemical-Protein Interactions (STITCH)
|   Direct binding, off-target effects, interaction confidence
|
+-- PHASE 7: Structural Alerts (ChEMBL)
|   PAINS, Brenk, Glaxo structural alerts
|
+-- SYNTHESIS: Integrated Risk Assessment
    Risk classification, evidence summary, data gaps, recommendations

See phase-procedures-detailed.md for complete tool parameters, decision logic, output templates, and fallback strategies for each phase.

---

Tool Summary by Phase

Phase 0: Compound Disambiguation

• PubChem_get_CID_by_compound_name (name: str)
• PubChem_get_compound_properties_by_CID (cid: int)
• ChEMBL_get_molecule (if ChEMBL ID available)

Phase 1: Predictive Toxicology

Dependency: ADMET-AI tools require pip install tooluniverse[ml]. If unavailable, skip to Phase 3 and use CTD + PubChemTox as alternatives.

• ADMETAI_predict_toxicity (smiles: list[str]) - AMES, DILI, ClinTox, LD50, hERG, etc.
• ADMETAI_predict_stress_response (smiles: list[str])
• ADMETAI_predict_nuclear_receptor_activity (smiles: list[str])

Phase 2: ADMET Properties

• ADMETAI_predict_BBB_penetrance / _bioavailability / _clearance_distribution / _CYP_interactions / _physicochemical_properties / _solubility_lipophilicity_hydration (all take smiles: list[str])

Phase 3: Toxicogenomics

• CTD_get_chemical_gene_interactions (input_terms: str) — chemical name, returns gene interactions across species
• CTD_get_chemical_diseases (input_terms: str) — chemical-disease associations with evidence type

Phase 3.5: PubChem Toxicity Data

• PubChemTox_get_toxicity_values (cid: int) — LD50, LC50, NOAEL reference values
• PubChemTox_get_ghs_classification (cid: int) — GHS hazard classification and pictograms
• PubChemTox_get_carcinogen_classification (cid: int) — NTP/IARC carcinogenicity assessments
• PubChemTox_get_acute_effects (cid: int) — acute toxicity by route/species
• PubChemTox_get_toxicity_summary (cid: int) — integrated toxicity overview

Phase 3.6: Adverse Outcome Pathways

• AOPWiki_list_aops (keyword: str) — search for relevant AOPs by chemical/mechanism
• AOPWiki_get_aop (aop_id: int) — full AOP detail: MIE, key events, adverse outcome

Phase 3.7: Environmental Exposure Context (US facilities)

Use for exposure/environmental-justice screening — locate regulated facilities near a community before assessing population-level exposure.

• EPA_search_tri_facilities (state, city, limit) — Toxics Release Inventory facilities reporting toxic chemical releases
• EPA_search_frs_facilities (state, city, limit) — Facility Registry Service (all EPA-regulated facilities) for broader siting/permitting context

Phase 4: Regulatory Safety (for pharmaceuticals only)

Environmental chemicals: Skip Phases 4-5 (no FDA labels/DrugBank). Use CTD + PubChemTox + AOPWiki instead.

• FDA_get_boxed_warning_info_by_drug_name / _contraindications_ / _adverse_reactions_ / _warnings_ (all take drug_name: str)

Phase 5: Drug Safety (for pharmaceuticals only)

• drugbank_get_safety_by_drug_name_or_drugbank_id (query, case_sensitive, exact_match, limit - all 4 required)

Phase 6: Chemical-Protein Interactions

• STITCH_get_chemical_protein_interactions (identifiers: list[str], species: int)
• Fallback (if STITCH fails for industrial chemicals): STRING_get_interaction_partners for key target genes (e.g., ESR1 for endocrine disruptors)
• DGIdb_get_drug_gene_interactions (genes: list[str]) — for target druggability context

Phase 7: Structural Alerts

• ChEMBL_search_compound_structural_alerts (molecule_chembl_id: str)

---

Risk Classification Matrix

---

Report Structure

# Chemical Safety & Toxicology Report: [Compound Name]
**Generated**: YYYY-MM-DD | **SMILES**: [...] | **CID**: [...]

## Executive Summary (risk classification + key findings, all graded)
## 1. Compound Identity (disambiguation table)
## 2. Predictive Toxicology (ADMET-AI endpoints)
## 3. ADMET Profile (absorption, distribution, metabolism, excretion)
## 4. Toxicogenomics (CTD chemical-gene-disease)
## 5. Regulatory Safety (FDA label data)
## 6. Drug Safety Profile (DrugBank)
## 7. Chemical-Protein Interactions (STITCH network)
## 8. Structural Alerts (ChEMBL)
## 9. Integrated Risk Assessment (classification, evidence summary, gaps, recommendations)
## Appendix: Methods and Data Sources

See report-templates.md for full section templates with example tables.

---

Mandatory Completeness Checklist

• [ ] Phase 0: Compound disambiguated (SMILES + CID minimum)
• [ ] Phase 1: At least 5 toxicity endpoints or "prediction unavailable"
• [ ] Phase 2: ADMET A/D/M/E sections or "not available"
• [ ] Phase 3: CTD queried; results or "no data in CTD"
• [ ] Phase 4: FDA labels queried; results or "not FDA-approved"
• [ ] Phase 5: DrugBank queried; results or "not found"
• [ ] Phase 6: STITCH queried; results or "no data available"
• [ ] Phase 7: Structural alerts checked or "ChEMBL ID not available"
• [ ] Synthesis: Risk classification with evidence summary
• [ ] Evidence Grading: All findings have [T1]-[T4] annotations
• [ ] Data Gaps: Explicitly listed

---

Common Use Patterns

1. Novel Compound: SMILES -> Phase 0 (resolve) -> Phase 1 (toxicity) -> Phase 2 (ADMET) -> Phase 7 (structural alerts) -> Synthesis 2. Approved Drug Review: Drug name -> All phases (0-7) -> Complete safety dossier 3. Environmental Chemical: Chemical name -> Phase 0 -> Phase 1-2 -> Phase 3 (CTD, key) -> Phase 6 (STITCH) -> Synthesis 4. Batch Screening: Multiple SMILES -> Phase 0 -> Phase 1-2 (batch) -> Comparative table -> Synthesis 5. Toxicogenomic Deep-Dive: Chemical + gene/disease interest -> Phase 0 -> Phase 3 (expanded CTD) -> Literature -> Synthesis

---

Limitations

• ADMET-AI: Computational [T3]; should not replace experimental testing
• CTD: May lag behind latest literature by 6-12 months
• FDA: Only covers FDA-approved drugs; not applicable to environmental chemicals
• DrugBank: Primarily drugs; limited industrial chemical coverage
• STITCH: Lower score thresholds increase false positives
• ChEMBL: Structural alerts require ChEMBL ID; not all compounds have one
• Novel compounds: May only have ADMET-AI predictions (no database evidence)
• SMILES validity: Invalid SMILES cause ADMET-AI failures

---

Reference Files

• phase-procedures-detailed.md - Complete tool parameters, decision logic, output templates, fallback strategies per phase
• evidence-grading.md - Evidence grading details and examples
• report-templates.md - Full report section templates with example tables
• phase-details.md - Additional phase context
• test_skill.py - Test suite

---

Summary

Total tools integrated: 25+ tools across 6 databases (ADMET-AI, CTD, FDA, DrugBank, STITCH, ChEMBL)

Best for: Drug safety assessment, chemical hazard profiling, environmental toxicology, ADMET characterization, toxicogenomic analysis

Outputs: Structured markdown report with risk classification (Critical/High/Medium/Low), evidence grading [T1-T4], and actionable recommendations

Evidence Grading System

Grade every toxicity claim by evidence strength.

Tier Definitions

Required Evidence Grading Locations

Evidence grades MUST appear in: 1. Executive Summary - Key toxicity findings graded 2. Toxicity Predictions - Every ADMET-AI endpoint with confidence note 3. Regulatory Safety - FDA findings marked [T1] 4. Chemical-Gene Interactions - CTD data marked by curation status 5. Risk Assessment - Final risk classification with supporting evidence tiers

Risk Classification Matrix

Phase Details: Chemical Safety Assessment

Phase 0: Compound Disambiguation (ALWAYS FIRST)

CRITICAL: Resolve compound identity before any analysis.

Input Types Handled

Resolution Steps

1. Input detection: Determine if input is name, SMILES, CID, or ChEMBL ID

• SMILES: contains typical SMILES characters (=, #, [, ], (, ), c, n, o and no spaces in middle)
• CID: numeric only
• ChEMBL: starts with "CHEMBL"
• Otherwise: treat as compound name

2. Name to CID: PubChem_get_CID_by_compound_name(name=<compound_name>) 3. CID to properties: PubChem_get_compound_properties_by_CID(cid=<cid>) 4. Extract SMILES: Get SMILES from PubChem properties (field: ConnectivitySMILES, CanonicalSMILES, or IsomericSMILES) 5. Store resolved IDs: Maintain dict with name, smiles, cid, formula, weight, inchi

---

Phase 1: Predictive Toxicology (ADMET-AI)

When: SMILES is available

Workflow

1. Call all 3 tools with smiles=[resolved_smiles] 2. Classification endpoints: Active (1) = toxic signal, Inactive (0) = no signal 3. Regression endpoints (LD50): Report numerical value with context 4. All predictions graded [T3]

Decision Logic

• hERG = Active: Flag prominently (cardiac safety risk)
• AMES = Active: Flag prominently (mutagenicity concern)
• DILI = Active: Flag prominently (liver toxicity concern)
• Multiple SMILES: Can batch up to ~10 SMILES in single call
• Failed prediction: Note "prediction unavailable" (don't fail entire report)

---

Phase 2: ADMET Properties

When: SMILES is available

Workflow

1. Call all 6 ADMET tools in parallel 2. Compile into Absorption / Distribution / Metabolism / Excretion sections 3. Assess Lipinski Rule of 5 compliance 4. Flag drug-drug interaction risks from CYP inhibition profiles

Decision Logic

• BBB penetrant + toxicity: Flag as neurotoxicity risk
• Low bioavailability: Note absorption concerns
• CYP3A4 inhibitor: Flag high DDI risk
• Lipinski violations: Report drug-likeness assessment

---

Phase 3: Toxicogenomics (CTD)

When: Compound name is resolved

Workflow

1. Call both CTD tools with input_terms=compound_name 2. Parse gene interactions: extract gene symbols, interaction types 3. Parse disease associations: extract disease names, evidence types (marker/mechanism/therapeutic) 4. Grade curated as [T2], inferred as [T3]

Decision Logic

• Direct vs inferred: CTD separates curated direct evidence from inferred associations
• Therapeutic vs toxic: Disease associations can be therapeutic or adverse
• Prioritize marker/mechanism: Stronger causal evidence than simple associations

---

Phase 4: Regulatory Safety (FDA Labels)

When: Compound has an approved drug name

Workflow

1. Call all 6 FDA tools in parallel 2. Prioritize: Boxed Warnings > Contraindications > Warnings > Adverse Reactions 3. All FDA label data is [T1] evidence 4. Boxed warning present: Flag as CRITICAL in executive summary 5. No FDA data: Note "Not an FDA-approved drug" and continue

---

Phase 5: Drug Safety Profile (DrugBank)

When: Compound is a known drug

Workflow

1. Call with query=drug_name, case_sensitive=False, exact_match=False, limit=5 2. Parse toxicity information, overdose data, contraindications 3. Cross-reference with FDA data; if conflict, defer to FDA [T1]

---

Phase 6: Chemical-Protein Interactions (STITCH)

When: Compound can be identified by name or SMILES

Workflow

1. Resolve compound: STITCH_resolve_identifier(identifier=compound_name, species=9606) 2. Get interactions with required_score=700 3. Flag safety-relevant targets: hERG (cardiac), CYP enzymes (metabolism), nuclear receptors (endocrine)

Confidence Levels

• >900: Well-established interaction [T2]
• 700-900: Probable interaction [T3]
• 400-700: Possible interaction, needs validation [T4]

---

Phase 7: Structural Alerts (ChEMBL)

When: ChEMBL molecule ID is available

Workflow

1. Call with molecule_chembl_id=chembl_id, limit=20 2. Parse alert types: PAINS, Brenk, Glaxo 3. No ChEMBL ID: Skip gracefully; note "structural alert analysis not available"

Chemical Safety: Detailed Phase Procedures

Referenced from SKILL.md. Contains detailed tool parameters, output templates, decision logic, and fallback strategies.

---

Phase 0: Compound Disambiguation (ALWAYS FIRST)

Input Types Handled

Resolution Steps

1. Input detection: Determine if input is name, SMILES, CID, or ChEMBL ID

• SMILES: contains typical SMILES characters (=, #, [, ], (, ), c, n, o and no spaces in middle)
• CID: numeric only
• ChEMBL: starts with "CHEMBL"
• Otherwise: treat as compound name

2. Name to CID: PubChem_get_CID_by_compound_name(name=<compound_name>) 3. CID to properties: PubChem_get_compound_properties_by_CID(cid=<cid>) 4. Extract SMILES: Get from PubChem properties (ConnectivitySMILES, CanonicalSMILES, or IsomericSMILES) 5. Store resolved IDs: Maintain dict with name, smiles, cid, formula, weight, inchi

Output Table

## Compound Identity
| Property | Value |
|----------|-------|
| **Name** | Acetaminophen |
| **PubChem CID** | 1983 |
| **SMILES** | CC(=O)Nc1ccc(O)cc1 |
| **Formula** | C8H9NO2 |
| **Molecular Weight** | 151.16 |

---

Phase 1: Predictive Toxicology (ADMET-AI)

When: SMILES is available

Tools (all take smiles: list[str])

Decision Logic

• Multiple SMILES: Can batch up to ~10 SMILES in single call
• Failed prediction: Note "prediction unavailable" (don't fail entire report)
• Confidence: All AI predictions are [T3], not definitive
• Critical flags: hERG Active = cardiac risk, AMES Active = mutagenicity, DILI Active = liver toxicity

Output Table

### Toxicity Predictions [T3]
| Endpoint | Prediction | Interpretation | Concern Level |
|----------|-----------|---------------|---------------|
| AMES Mutagenicity | Inactive | No mutagenic signal | Low |
| DILI | Active | Drug-induced liver injury risk | HIGH |
| hERG Inhibition | Active | Cardiac arrhythmia risk | HIGH |
*All predictions from ADMET-AI. Evidence tier: [T3]*

---

Phase 2: ADMET Properties

When: SMILES is available

Tools (all take smiles: list[str])

Decision Logic

• BBB penetrant + toxicity: Flag as neurotoxicity risk
• Low bioavailability: F20% = Low -> absorption concerns
• CYP inhibitor: CYP3A4 inhibitor = Yes -> high DDI risk
• Lipinski violations: Count and report drug-likeness assessment

Output Format

### ADMET Profile [T3]
#### Absorption
| Property | Value | Interpretation |
|----------|-------|----------------|
| BBB Penetrance | Yes | Crosses blood-brain barrier |
| Bioavailability (F20%) | 85% | Good oral absorption |

#### Metabolism
| CYP Enzyme | Substrate | Inhibitor |
|------------|-----------|-----------|
| CYP3A4 | Yes | Yes (DDI risk) |

---

Phase 3: Toxicogenomics (CTD)

When: Compound name is resolved

Tools

Decision Logic

• Direct evidence vs inferred: CTD separates curated (grade [T2]) from inferred ([T3])
• Therapeutic vs toxic: Disease associations can be therapeutic or adverse
• Prioritize marker/mechanism: Stronger causal evidence than simple associations

---

Phase 4: Regulatory Safety (FDA Labels)

When: Compound has an approved drug name

Tools (all take drug_name: str)

Decision Logic

• Boxed warning present: Flag as CRITICAL in executive summary
• No FDA data: Note "Not an FDA-approved drug" and continue
• All FDA findings: Grade as [T1]; nonclinical toxicology as [T2]

---

Phase 5: Drug Safety Profile (DrugBank)

When: Compound is a known drug

Tool

drugbank_get_safety_by_drug_name_or_drugbank_id(query=drug_name, case_sensitive=False, exact_match=False, limit=5)

All 4 parameters required. Cross-reference with FDA data; defer to FDA [T1] on conflicts.

---

Phase 6: Chemical-Protein Interactions (STITCH)

Tools

Confidence Scoring

• >900: Well-established [T2]
• 700-900: Probable [T3]
• 400-700: Possible, needs validation [T4]
• Flag safety-relevant targets: hERG (cardiac), CYP enzymes (metabolism), nuclear receptors (endocrine)

---

Phase 7: Structural Alerts (ChEMBL)

When: ChEMBL molecule ID is available

Tool

ChEMBL_search_compound_structural_alerts(molecule_chembl_id=chembl_id, limit=20)

Alert types: PAINS (pan-assay interference), Brenk (medicinal chemistry), Glaxo (GSK). No ChEMBL ID -> skip gracefully.

---

Fallback Strategies

Compound Resolution

• Primary: PubChem by name -> CID -> properties -> SMILES
• Fallback 1: ChEMBL search by name -> molecule -> SMILES
• Fallback 2: If SMILES provided directly, skip name resolution

Toxicity Prediction

• Primary: All 9 ADMET-AI endpoints
• Fallback: If fails, note "prediction failed" and continue with database evidence

Regulatory Data

• Primary: FDA labels by drug name
• Fallback: Try alternative names (brand vs generic)

CTD Data

• Primary: Search by common chemical name
• Fallback: Try MeSH name

---

Tool Parameter Quick Reference

Response Format Notes

• ADMET-AI: {status: "success", data: {...}}
• CTD: List of interaction/association objects
• FDA: {status, data} with label text
• DrugBank: {data: [...]} with drug records
• STITCH: List of interaction objects with scores
• PubChem CID lookup: {IdentifierList: {CID: [...]}}
• PubChem properties: Dict with CID, MolecularWeight, ConnectivitySMILES, IUPACName

Report Templates: Chemical Safety Assessment

Output Report Structure

All analyses generate a structured markdown report:

# Chemical Safety & Toxicology Report: [Compound Name]

**Generated**: YYYY-MM-DD HH:MM
**Compound**: [Name] | SMILES: [SMILES] | CID: [CID]

## Executive Summary
[2-3 sentence overview with risk classification and key findings, all graded]

## 1. Compound Identity
[Phase 0 results - disambiguation table]

## 2. Predictive Toxicology
[Phase 1 results - ADMET-AI toxicity endpoints]

## 3. ADMET Profile
[Phase 2 results - absorption, distribution, metabolism, excretion]

## 4. Toxicogenomics
[Phase 3 results - CTD chemical-gene-disease relationships]

## 5. Regulatory Safety
[Phase 4 results - FDA label information]

## 6. Drug Safety Profile
[Phase 5 results - DrugBank data]

## 7. Chemical-Protein Interactions
[Phase 6 results - STITCH network]

## 8. Structural Alerts
[Phase 7 results - ChEMBL alerts]

## 9. Integrated Risk Assessment
[Synthesis - risk classification, evidence summary, data gaps, recommendations]

## Appendix: Methods and Data Sources
[Tool versions, databases queried, date of access]

---

Completeness Checklist

Before finalizing any report, verify:

• [ ] Phase 0: Compound fully disambiguated (SMILES + CID at minimum)
• [ ] Phase 1: At least 5 toxicity endpoints reported or "prediction unavailable" noted
• [ ] Phase 2: ADMET profile with A/D/M/E sections or "not available" noted
• [ ] Phase 3: CTD queried; gene interactions and disease associations reported or "no data in CTD"
• [ ] Phase 4: FDA labels queried; results or "not an FDA-approved drug" noted
• [ ] Phase 5: DrugBank queried; results or "not found in DrugBank" noted
• [ ] Phase 6: STITCH queried; results or "no STITCH data available" noted
• [ ] Phase 7: Structural alerts checked or "ChEMBL ID not available" noted
• [ ] Synthesis: Risk classification provided with evidence summary
• [ ] Evidence Grading: All findings have [T1]-[T4] annotations
• [ ] Data Gaps: Explicitly listed in synthesis section

---

Example Tables

Compound Identity

| Property | Value |
|----------|-------|
| **Name** | Acetaminophen |
| **PubChem CID** | 1983 |
| **SMILES** | CC(=O)Nc1ccc(O)cc1 |
| **Formula** | C8H9NO2 |
| **Molecular Weight** | 151.16 |

Toxicity Predictions [T3]

| Endpoint | Prediction | Interpretation | Concern Level |
|----------|-----------|---------------|---------------|
| AMES Mutagenicity | Inactive | No mutagenic signal | Low |
| ClinTox | Active | Clinical toxicity signal | HIGH |
| DILI | Active | Drug-induced liver injury risk | HIGH |
| LD50 (Zhu) | 2.45 log(mg/kg) | ~282 mg/kg (moderate) | Medium |
| hERG Inhibition | Active | Cardiac arrhythmia risk | HIGH |

ADMET Profile

#### Absorption
| Property | Value | Interpretation |
|----------|-------|----------------|
| BBB Penetrance | Yes | Crosses blood-brain barrier |
| Bioavailability (F20%) | 85% | Good oral absorption |

#### Metabolism
| CYP Enzyme | Substrate | Inhibitor |
|------------|-----------|-----------|
| CYP1A2 | No | No |
| CYP3A4 | Yes | Yes (DDI risk) |

Integrated Risk Assessment

### Overall Risk Classification: [HIGH]

| Dimension | Finding | Evidence Tier | Concern |
|-----------|---------|--------------|---------|
| ADMET Toxicity | DILI active, hERG active | [T3] | HIGH |
| FDA Label | Boxed warning for hepatotoxicity | [T1] | CRITICAL |
| CTD Toxicogenomics | 156 gene interactions | [T2] | HIGH |

### Key Safety Concerns
1. **Hepatotoxicity** [T1]: FDA boxed warning + ADMET-AI DILI + CTD liver associations
2. **Cardiac Risk** [T3]: hERG prediction + STITCH hERG interaction

### Data Gaps
- [ ] No in vivo genotoxicity data
- [ ] STITCH scores moderate (700-900)

### Recommendations
1. Avoid doses >4g/day [T1]
2. Monitor liver function in chronic use [T1]
3. Screen for CYP3A4 interactions [T3]

---

Common Use Patterns