KODA: An Agentic Framework for KEGG Orthology-Driven Discovery of Antimicrobial Drug Targets in Gut Microbiome

We introduce KODA, a multi-agent framework that combines LLMs with a Neo4j-based knowledge graph to identify potential antimicrobial drug targets in the human gut microbiome. KODA enables natural language querying of microbiome data and generates analytical reports focused on KEGG orthologies (KOs) linked to essential microbial genes.

System Highlights

• Agents: Specialized LLM agents collaborate to process queries, generate Cypher, and create reports
• Graph Backend: A structured Neo4j graph links microbes, genes, pathways, and interactions
• Evaluator: LLM-based reviewers assess the relevance and correctness of outputs

Installation

Create a virtual environment and install dependencies:

python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows

pip install -r requirements.txt

Requirements

Key dependencies include:

• crewai==0.19.0
• langchain==0.1.14
• openai==1.22.0
• neo4j==5.19.0
• sentence-transformers==2.7.0
• transformers==4.51.3
  (See full list in requirements.txt)

Usage

1. Set up and populate the Neo4j microbiome knowledge graph
2. Add your OpenAI and Neo4j credentials to a .env file
3. Run the main script:

python main.py

4. Enter queries like:
   "List essential KOs involved in short-chain fatty acid production"

Citation

If you use this repository, please cite:

KODA: An Agentic Framework for KEGG Orthology-Driven Discovery of Antimicrobial Drug Targets in Gut Microbiome

License

MIT License – see the LICENSE file for details.