Genentech gRED Research Agent: AI-Powered Drug Target Identification with Amazon Bedrock

Title

Genentech gRED Research Agent: AI-Powered Drug Target Identification with Amazon Bedrock

Content

Genentech partnered with Amazon Web Services (AWS) to develop a solution that automates the time-consuming manual search process for drug target identification and biomarker validation. The company designed a generative AI system called gRED Research Agent, built using Anthropic Claude Sonnet 3.5 in Amazon Bedrock Agents. The system can process complex scientific queries like "What cell surface receptors are enriched in specific cells in inflammatory bowel disease?" and simultaneously search through multiple data sources from PubMed journals to internal repositories, synthesizing findings with cited summaries. The agent uses autonomous agents that break down complicated research tasks into dynamic, multi-step workflows. Unlike traditional automation systems that follow predetermined paths, these agents adapt their approach based on information gathered at each step, access and analyze multiple knowledge bases using Retrieval Augmented Generation (RAG), and execute complex queries by interfacing with Genentech's internal APIs and databases. The system transforms processes that once took scientists weeks to accomplish into tasks that take just minutes. This is expected to help Genentech automate over 43,000 hours of manual effort in biomarker validation across therapeutic areas, reducing time-to-target identification. John Marioni, Senior Vice President and Head of Computational Research at Genentech, noted that the system democratizes access to data sets and computational tools for scientists with less computational background. Aviv Regev, Executive Vice President and Head of Genentech Research and Early Development, emphasized that agents boost scientists' capabilities rather than replacing them. Genentech is further developing a network of specialized sub-agents through Amazon Bedrock's multi-agent collaboration capabilities, with dedicated agents for specific domains such as public data sources like PubMed and internal proprietary databases.