Skip to content

QUICK_START.md

Latest commit

 

History

History
147 lines (105 loc) · 5.15 KB

QUICK_START.md

File metadata and controls

147 lines (105 loc) · 5.15 KB

Chemical Safety & Toxicology - Quick Start

Overview

This skill performs comprehensive chemical safety assessment using 25+ tools across 6 databases (ADMET-AI, CTD, FDA, DrugBank, STITCH, ChEMBL). It generates structured reports with evidence grading and risk classification.

Quick Start Examples

Example 1: Assess Safety of a Known Drug

Query: "Assess the toxicity profile of Acetaminophen"

What happens:

  1. Resolves "Acetaminophen" to SMILES via PubChem (CID: 1983)
  2. Runs 9 ADMET-AI predictions (toxicity, BBB, bioavailability, CYP, etc.)
  3. Queries CTD for chemical-gene and chemical-disease associations
  4. Extracts FDA label safety data (boxed warnings, contraindications, adverse reactions)
  5. Retrieves DrugBank safety profile
  6. Maps STITCH chemical-protein interactions
  7. Checks ChEMBL structural alerts
  8. Generates integrated risk assessment

Output: Acetaminophen_safety_report.md with risk classification and recommendations

Example 2: Screen Novel Compound by SMILES

Query: "Predict toxicity for this compound: CC(=O)Oc1ccccc1C(=O)O"

What happens:

  1. Detects SMILES input, resolves to PubChem CID (Aspirin, CID: 2244)
  2. Runs full ADMET-AI toxicity + ADMET property prediction
  3. Queries CTD, FDA, DrugBank for known safety data
  4. Generates comparative predictive vs. known safety profile

Example 3: Environmental Chemical Risk Assessment

Query: "What are the health risks of Bisphenol A?"

What happens:

  1. Resolves "Bisphenol A" via PubChem
  2. Runs ADMET-AI predictions
  3. Queries CTD extensively (BPA has rich toxicogenomics data)
  4. FDA/DrugBank phases skipped (not an approved drug)
  5. Maps STITCH interactions (endocrine disruption targets)
  6. Generates environmental health risk report

Example 4: Batch Toxicity Screening

Query: "Compare the toxicity profiles of ibuprofen, naproxen, and celecoxib"

What happens:

  1. Resolves all three names to SMILES
  2. Runs ADMET-AI batch predictions
  3. Creates comparative toxicity table
  4. Queries safety data for each compound
  5. Generates comparative risk assessment

Python SDK Usage

from tooluniverse import ToolUniverse

tu = ToolUniverse()
tu.load_tools()

# Phase 0: Resolve compound
cid_result = tu.tools.PubChem_get_CID_by_compound_name(name="Acetaminophen")
cid = cid_result['data']['IdentifierList']['CID'][0]  # 1983

props = tu.tools.PubChem_get_compound_properties_by_CID(cid=cid)
smiles = props['data']['PropertyTable']['Properties'][0]['CanonicalSMILES']

# Phase 1: Toxicity predictions
tox = tu.tools.ADMETAI_predict_toxicity(smiles=[smiles])
print("Toxicity:", tox)

# Phase 2: ADMET
bbb = tu.tools.ADMETAI_predict_BBB_penetrance(smiles=[smiles])
cyp = tu.tools.ADMETAI_predict_CYP_interactions(smiles=[smiles])
print("BBB:", bbb)
print("CYP:", cyp)

# Phase 3: Toxicogenomics
gene_interactions = tu.tools.CTD_get_chemical_gene_interactions(input_terms="Acetaminophen")
disease_assoc = tu.tools.CTD_get_chemical_diseases(input_terms="Acetaminophen")

# Phase 4: FDA Safety
boxed = tu.tools.FDA_get_boxed_warning_info_by_drug_name(drug_name="Acetaminophen")
adverse = tu.tools.FDA_get_adverse_reactions_by_drug_name(drug_name="Acetaminophen")

# Phase 5: DrugBank Safety
safety = tu.tools.drugbank_get_safety_by_drug_name_or_drugbank_id(
    query="Acetaminophen", case_sensitive=False, exact_match=False, limit=5
)

MCP Integration

When used via MCP (Claude Desktop, Cursor, etc.), simply ask:

  • "Assess the toxicity of metformin"
  • "Is caffeine safe? Generate a comprehensive safety report"
  • "Predict ADMET properties for SMILES: c1ccc(cc1)O"
  • "What genes does arsenic interact with?"
  • "Compare safety profiles of aspirin and ibuprofen"

The skill will automatically invoke the appropriate tools and generate a report.

Key Tool Reference

Tool Input What It Returns
ADMETAI_predict_toxicity SMILES list AMES, DILI, LD50, hERG, carcinogenicity predictions
ADMETAI_predict_BBB_penetrance SMILES list BBB crossing probability
ADMETAI_predict_CYP_interactions SMILES list CYP1A2/2C9/2C19/2D6/3A4 inhibition/substrate
CTD_get_chemical_gene_interactions Chemical name Gene interaction list with types
CTD_get_chemical_diseases Chemical name Disease association list
FDA_get_boxed_warning_info_by_drug_name Drug name Black box warnings
FDA_get_adverse_reactions_by_drug_name Drug name Known adverse reactions
drugbank_get_safety_by_drug_name_or_drugbank_id Query + flags Toxicity, contraindications
STITCH_get_chemical_protein_interactions STITCH IDs Chemical-protein interaction network
ChEMBL_search_compound_structural_alerts ChEMBL ID Structural toxicity flags

Evidence Tiers

Tier Meaning Source Examples
[T1] Direct human/regulatory evidence FDA labels, clinical trial data
[T2] Animal studies, curated databases CTD curated, DrugBank, nonclinical toxicology
[T3] Computational prediction, association ADMET-AI, CTD inferred, STITCH scores
[T4] Database annotation, text-mined Literature mentions, unvalidated predictions