Pharmaclaw Market Intel Agent

Fetches and analyzes FAERS (FDA Adverse Event Reporting System) data from openFDA API. Supports drug names and SMILES (resolves via PubChem). Generates: events list, yearly trends (counts), top reactions/outcomes as JSON + matplotlib bar chart PNGs.

Pharma Market Intel Agent - FAERS Query Skill

Overview

Query real-world post-market safety data for drugs. Useful for market intel on safety profiles, emerging risks, competitor analysis.

Key outputs:

• JSON summaries (trends, top reactions/outcomes)
• PNG bar charts (yearly reports, top 10 reactions/outcomes)
• Sample recent events

Rate limits: openFDA ~240 req/min. Counts are fast (no full data).

Chemistry-Query Structure

Parse user queries into this model for standardized chaining:

python

from dataclasses import dataclass
from typing import List, Optional

@dataclass
class ChemistryQuery:
    drug: str  # Drug name or SMILES
    query_type: str = 'faers'  # 'faers', 'pubchem', etc.
    metrics: Optional[List[str]] = None  # ['yearly_trends', 'top_reactions', 'top_outcomes', 'events']
    limit_events: int = 20

Example:

json

{
  \"drug\": \"aspirin\",  // or \"CC(=O)OC1=CC=CC=C1C(=O)O\"
  \"query_type\": \"faers\",
  \"metrics\": [\"yearly_trends\", \"top_reactions\"]
}

Quick Start / Workflows

1. Basic Query (All Metrics)

text

exec skills/pharma-market-intel-agent/scripts/query_faers.py --drug aspirin --output ./aspirin_faers

Generates:

• aspirin_faers/aspirin_summary.json
• *.png plots
• Recent events JSON

2. SMILES Input

text

exec ... --drug \"CC(=O)OC1=CC=CC=C1C(=O)O\"  # Aspirin SMILES

Auto-resolves to name via PubChem.

3. Custom Limit

text

exec ... --drug ozempic --limit-events 50 --output ozempic_analysis

Chaining Examples

• With chemistry-query: Resolve/validate SMILES first, then FAERS.
• pharma-tox-agent: Feed top reactions for tox prediction.
• pharma-ip-expansion-agent: Check safety for IP expansion targets.
• traction-agent: Market risk scoring from FAERS trends.

text

# Agent workflow:
1. Parse ChemistryQuery
2. Resolve SMILES if needed (pubchempy or query_faers handles)
3. Run query_faers.py
4. Read PNGs/JSONs into response
5. Chain if metrics require

---

ClinicalTrials.gov Integration

Query clinical trial data from ClinicalTrials.gov API v2. Search by drug, condition, phase, and status. No API key needed.

Quick Start

bash

# Search by drug
exec skills/pharma-market-intel-agent/scripts/query_trials.py --drug "sotorasib" --output ./sotorasib_trials

# Search by condition + filters
exec ... --condition "breast cancer" --phase PHASE3 --status RECRUITING --limit 10 --output ./bc_trials

# Search by both
exec ... --drug "pembrolizumab" --condition "NSCLC" --output ./pembro_trials

# SMILES input (auto-resolves via PubChem)
exec ... --drug "CC(=O)OC1=CC=CC=C1C(=O)O" --output ./aspirin_trials

Outputs

• {drug}_trials_summary.json — Full structured summary with trials list and aggregate stats
• {drug}_trials_by_phase.png — Bar chart by phase
• {drug}_trials_by_status.png — Bar chart by status
• {drug}_trials_timeline.png — Timeline of trial start dates

JSON Summary Structure

json

{
  "drug": "sotorasib",
  "total_found": 45,
  "trials": [{"nct_id": "NCT...", "title": "...", "phase": "PHASE3", "sponsor": "Amgen", ...}],
  "stats": {"by_phase": {...}, "by_status": {...}, "top_sponsors": [...], "top_conditions": [...]}
}

Chaining Examples

• Chemistry Query → Market Intel: Resolve SMILES, then query trials for competitive landscape
• FAERS + Trials: Run both scripts for a drug to get safety profile + development pipeline
• chain_entry.py: Use --metrics trials or --metrics faers,trials to run both in one call

---

References

• faers_fields.md: Key FAERS fields & search syntax
• clinicaltrials_fields.md: ClinicalTrials.gov API fields & enums
• openFDA Drug Event API
• ClinicalTrials.gov API v2
• PubChem PUG REST

Resources

• scripts/query_faers.py: FAERS query executable
• scripts/query_trials.py: ClinicalTrials.gov query executable
• scripts/chain_entry.py: Unified entry point (faers + trials)
• assets/: Store generated PNGs here for reuse