Accelerating Drug Discovery with AI: Building a Research Assistant Using Strands Agents and Amazon Bedrock

Overview of AI in Life Sciences and Drug Discovery

Understanding Strands Agents for Advanced Applications

Creating a Drug Discovery Research Assistant: Step-by-Step Guide

Prerequisites for Implementation

Connecting to Foundation Models

Utilizing Model Context Protocol Tools

Developing Specialized Sub-Agents for Enhanced Functionality

Orchestrating the Research Workflow

Real-World Application: Exploring Breast Cancer Research

Cleanup Procedures After Deployment

Conclusion: The Future of AI in Drug Discovery

About the Authors

Revolutionizing Drug Discovery with AI: Building a Research Assistant Using Strands Agents and Amazon Bedrock

The world of drug discovery is at a pivotal point, where advanced artificial intelligence transforms the complexities of scientific research into streamlined, efficient workflows. Traditional methods of navigating vast scientific literature, clinical trial data, and molecular databases are cumbersome and time-consuming. Recognizing this, leading life science companies like Genentech and AstraZeneca are increasingly leveraging AI agents and generative AI tools to expedite scientific discovery. Utilizing platforms such as Amazon Bedrock has enabled rapid deployment of domain-specific workflows, unlocking possibilities from early drug target identification to enhanced healthcare provider engagement.

In this blog post, we’ll explore how to harness the power of Strands Agents and Amazon Bedrock to create an intelligent research assistant that revolutionizes the drug discovery process.

Solution Overview

The proposed solution demonstrates how Strands Agents can connect high-performance foundation models (FMs) to widely-used life science data sources like arXiv, PubMed, and ChEMBL. It also employs the Model Context Protocol (MCP) to query multiple databases simultaneously, allowing the AI assistant to synthesize its findings into comprehensive reports on drug targets, disease mechanisms, and therapeutic areas.

The Orchestrator Agent

This solution employs a team of small, focused AI agents harmonizing their efforts through an orchestrator agent. This orchestrator is fundamental to managing user queries and routing them to specialized sub-agents equipped for either information retrieval or report generation.

Prerequisites

Before diving into the implementation, make sure you have the following prerequisites ready:

Python 3.10 or higher
Strands Agents package
Additional Python dependencies
AWS account with access to the required Amazon Bedrock Foundation Models:
- Anthropic’s Claude 3.7 Sonnet
- Anthropic’s Claude 3.5 Sonnet
- Anthropic’s Claude 3.5 Haiku

To set up your development environment, follow these steps:

Clone the code repository from GitHub.
Install the required Python dependencies with pip install -r requirements.txt.
Configure your AWS credentials as environment variables or through a credentials file.
Save your Tavily API key in a .env file.

Defining the Foundation Model

The next step involves establishing a connection to an FM within Amazon Bedrock using the BedrockModel class from Strands Agents. Below is a sample code snippet to set this up:

from strands import Agent, tool
from strands.models import BedrockModel

def get_model():
    model = BedrockModel(
        boto_client_config=Config(
            read_timeout=900,
            connect_timeout=900,
            retries=dict(max_attempts=3, mode="adaptive"),
        ),
        model_id="us.anthropic.claude-3-7-sonnet-20250219-v1:0",
        max_tokens=64000,
        temperature=0.1,
        top_p=0.9
    )
    return model

Implementing MCP Tools

Integrating various scientific databases into our solution is crucial for comprehensive research. We can utilize MCP servers to connect to several key life science tools, including:

arXiv: for scholarly articles
PubMed: for biomedical literature
ChEMBL: for bioactive molecules
ClinicalTrials.gov: for research studies
Tavily Web Search: for internet content

Here’s an example code snippet to set up the MCP clients for these tools:

tavily_mcp_client = MCPClient(lambda: stdio_client(
    StdioServerParameters(command="python", args=["application/mcp_server_tavily.py"])
))
# Similar setup for arXiv, PubMed, ChEMBL, and ClinicalTrials

Creating Specialized Sub-Agents

We can define specialized sub-agents for planning and synthesis tasks. For instance, the planning agent can analyze research questions to determine the most relevant databases to query:

@tool
def planning_agent(query: str) -> str:
    ...
    response = planner(planning_prompt)
    return str(response)

The synthesis agent will integrate findings into comprehensive reports:

@tool
def synthesis_agent(research_results: str) -> str:
    ...
    response = synthesis(synthesis_prompt)
    return str(response)

Developing the Orchestration Agent

Finally, we will create an orchestration agent that coordinates the entire workflow, maintaining conversation history and handling user queries. This agent will utilize the SlidingWindowConversationManager class to remember the last 10 exchanges.

def create_orchestrator_agent():
    ...
    orchestrator = Agent(
        model=model,
        system_prompt=system,
        tools=tools,
        conversation_manager=conversation_manager
    )
    return orchestrator

Example Use Case: Researching Breast Cancer

To illustrate the power of this AI assistant, we can test it by querying for a comprehensive report on HER2 in breast cancer research. The assistant will analyze recent news articles, relevant recent research, compounds, and ongoing clinical trials, ultimately synthesizing these results into a structured report.

Generated Report Excerpt:

Comprehensive Scientific Report: HER2 in Breast Cancer Research and Treatment

Executive Summary: Recent advancements in the understanding of HER2 biology highlight new therapeutic approaches, notably the emergence of antibody-drug conjugates (ADCs) that offer promising efficacy.

Conclusion

In this post, we showcased how Strands Agents can streamline the creation of domain-specific AI assistants tailored for drug discovery. As scientific information continues to proliferate, such frameworks will be essential for researchers to navigate complex data landscapes efficiently.

We encourage you to experiment with this solution, adapting it to your own research queries, and explore the extensive capabilities of Strands Agents and Amazon Bedrock.

For further insights on intelligent agents in healthcare and life sciences, don’t hesitate to visit our GitHub repository for more resources.

By implementing these advanced AI-driven tools, we can not only enhance the speed of drug discovery but also pave the way for breakthroughs that can significantly impact human health.

Exclusive Content:

Create a Drug Discovery Research Assistant with Strands Agents and Amazon Bedrock