Advanced Biomedical Agent

A comprehensive biomedical research agent that integrates multiple data sources (OpenFDA, DailyMed, ClinicalTrials.gov, ChEMBL, BindingDB, DrugCentral, Orange Book) into a unified PostgreSQL database with LLM-friendly search tools.

The Advanced Biomedical Agent provides:

• Unified Database: Integrates 9 major biomedical data sources into a single PostgreSQL database
• Intelligent Search Tools: Both retrieval-based and agent-based tools for querying biomedical data
• LLM Integration: Tools designed for use with language models, with automatic summarization and formatting
• Comprehensive Coverage: Clinical trials, drug labels, molecular structures, target interactions, adverse events, and more

1. OpenFDA - Drug labels with normalized structure
2. DailyMed - SPL drug labels with semantic search
3. ClinicalTrials.gov - Clinical trials from CTTI AACT dump
4. Orange Book - FDA therapeutic equivalence data
5. BindingDB - Molecular targets & binding affinity
6. ChEMBL - Biochemical annotations
7. DrugCentral - Molecular structures & chemical data
8. Target Mapping - Canonical target mapping (ChEMBL + BindingDB)
9. Molecule Mapping - Unified molecules + ClinicalTrials.gov/product mappings

Direct database access functions that return structured data:

• Clinical trials search with flexible filtering
• Drug labels search (OpenFDA + DailyMed)
• Molecule-trial connectivity
• Adverse events search
• Clinical trial outcomes
• Orange Book search
• Cross-database identifier lookup
• Biotherapeutic sequence search
• Target/drug pharmacology search

LLM-friendly wrappers with formatting and summarization:

• Automatic output summarization for long results
• Input normalization for robust LLM integration
• Helpful error messages with suggestions
• Research output management (store/retrieve full results)
• Web search capabilities
• Thinking/reasoning tool

• Clinical Trial Search: Search 500,000+ clinical trials with flexible querying
• Drug Information: Access FDA drug labels, indications, warnings, and interactions
• Pharmacology: Drug-target interactions, mechanisms of action, activity data (IC50, Ki, Kd)
• Molecular Search: Structure similarity, substructure search, exact structure matching
• Safety Data: Adverse events, outcomes, and safety comparisons
• Regulatory Data: Orange Book patents, exclusivity, and therapeutic equivalence
• Cross-Database Lookup: Resolve identifiers across multiple databases

• PostgreSQL 14+ (recommended: 16)
• Python 3.12+
• ~50 GB disk space for full database
• 8+ GB RAM recommended for ingestion

# Clone the repository
git clone <repository-url>
cd AdvancedBiomedicalAgent

# Install the package
pip install -e .

# Install optional dependencies
pip install -e ".[chem]"  # For RDKit support
pip install -e ".[dev]"   # For development

See the Setup Guide for detailed instructions.

# Quick setup
biomedagent-db setup-db
biomedagent-db install-extensions

See the Data Population Guide for detailed instructions.

# Full ingestion (all sources)
biomedagent-db ingest

# Selective ingestion
biomedagent-db ingest --skip-openfda --skip-ctgov

# Use local files (skip download)
biomedagent-db ingest --openfda-use-local-files

from bioagent.agent.tools import get_summarized_tools
from langchain_openai import ChatOpenAI

# Initialize LLM
llm = ChatOpenAI(model="gpt-4")

# Get tools with summarization
tools = get_summarized_tools(
    summarizer_llm=llm,
    session_id="my-session"
)

# Use tools in your agent
# Tools are automatically registered and ready to use

• Setup Guide - Complete database setup instructions
• Database Schema - Detailed database structure documentation
• Data Population Guide - How to ingest data from all sources
• Molecular Mapping Guide - Understanding molecular mappings
• Tools Documentation - Comprehensive tool reference

AdvancedBiomedicalAgent/
├── src/bioagent/
│   ├── agent/
│   │   └── tools/          # Agent-based tools (@tools)
│   │       ├── dbsearch.py
│   │       ├── target_search.py
│   │       ├── web_search.py
│   │       ├── summarizing.py
│   │       └── thinking.py
│   ├── data/
│   │   ├── search/         # Retrieval-based tools (@search)
│   │   │   ├── clinical_trial_search.py
│   │   │   ├── target_search.py
│   │   │   ├── openfda_and_dailymed_search.py
│   │   │   └── ...
│   │   └── ingest/         # Data ingestion scripts
│   └── cli/                 # Command-line interface
├── docs/                    # Documentation
├── notebooks/               # Jupyter notebooks
└── tests/                   # Test suite

The system uses a two-layer architecture:

1. Retrieval Tools (@search): Direct database access functions

   • Return structured Pydantic models
   • Async/await for performance
   • Located in src/bioagent/data/search/

2. Agent Tools (@tools): LLM-friendly wrappers

   • Format outputs for LLM consumption
   • Automatic summarization for long outputs
   • Input normalization and validation
   • Located in src/bioagent/agent/tools/

See Tools Documentation for complete details.

# Using agent tool
result = await search_clinical_trials(
    condition="non-small cell lung cancer",
    intervention="pembrolizumab",
    phase="Phase 3",
    status="Recruiting",
    limit=10
)

# Find all targets for a drug
result = await search_drug_targets(
    drug_name="imatinib",
    min_pchembl=6.0,
    data_source="both"
)

# Get drug label information
result = await search_drug_labels(
    drug_names="warfarin",
    section_queries=["warnings", "drug interactions"],
    fetch_all_sections=False
)

# Database management
biomedagent-db setup-db              # Create database
biomedagent-db install-extensions     # Install pgvector, RDKit
biomedagent-db verify-deps            # Verify dependencies

# Data ingestion
biomedagent-db ingest                 # Full ingestion
biomedagent-db ingest --skip-<source> # Skip specific source

# Database utilities
biomedagent-db vacuum                 # Optimize database
biomedagent-db stats                  # Show database statistics

# Install development dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black src/
ruff check src/

# Type checking
mypy src/

• PostgreSQL 14+ with pgvector and RDKit extensions
• Python 3.12+
• ~50 GB disk space for full database
• 8+ GB RAM for ingestion

See LICENSE file for details.

Contributions are welcome! Please see the documentation for:

• Database schema details
• Tool development guidelines
• Data ingestion processes

For issues and questions:

1. Check the documentation in docs/
2. Review the Setup Guide for common issues
3. Check the Database Schema for data structure details

Note: This project integrates large-scale biomedical datasets. Ensure you have adequate disk space and follow the setup guide carefully.