llm_usage.md

Using Reactome MCP with LLMs

This document demonstrates how Reactome MCP tools can be used within LLM-based workflows, such as agent systems and retrieval-augmented pipelines.

The goal is to illustrate how natural language queries can be mapped to MCP tool calls and how results can be composed into structured responses.

---

Example 1: Pathway Search

User query: What pathways are related to cancer?

LLM Action:

• Select tool: search_pathways
• Input: "cancer"

Expected Result:

• List of pathways related to cancer
• Pathway identifiers
• Descriptions and summaries

LLM Response (example): The following pathways are associated with cancer:

• Pathway A — Description...
• Pathway B — Description...

---

Example 2: Gene to Pathway Mapping

User query: Which pathways involve TP53?

LLM Action:

• Select tool: get_pathways_by_gene
• Input: "TP53"

Expected Result:

• Pathways involving TP53
• Associated biological processes

LLM Response (example): TP53 is involved in pathways related to cell cycle regulation, apoptosis, and DNA damage response.

---

Example 3: Multi-step Reasoning

User query: Compare pathways involved in cancer and apoptosis.

LLM Steps:

1. Call search_pathways("cancer")
2. Call search_pathways("apoptosis")
3. Aggregate and compare results

LLM Response (example): Cancer and apoptosis share overlapping pathways such as those related to cell survival and programmed cell death.

---

Example 4: LangChain Integration

Below is a minimal example showing how Reactome APIs can be wrapped as tools and used in an LLM agent.

from langchain.tools import tool
from langchain.agents import initialize_agent
from langchain.chat_models import ChatOpenAI
import requests

@tool
def search_pathways(query: str):
    """Search biological pathways in Reactome"""
    url = f"https://reactome.org/ContentService/search/query?q={query}"
    response = requests.get(url)
    return response.json()

llm = ChatOpenAI(temperature=0)

tools = [search_pathways]

agent = initialize_agent(
    tools,
    llm,
    agent="zero-shot-react-description",
    verbose=True
)

agent.run("What pathways are related to cancer?")