How a Singapore cancer institute automated genomics data operations on AWS

This article describes how the Cancer Science Institute of Singapore built VDOT, a Virtual DataOps Team of five specialized AI agents running on Amazon Bedrock AgentCore that automates genomics data operations through conversational interfaces.

• Researchers can discover, download, and process genomic datasets from fragmented public archives (GDC, SRA, EGA) through natural language requests
• Five specialized agents handle orchestration, repository-specific discovery, secure data transfer, workflow processing, and quality review
• Each discovery agent uses repository-specific APIs via Model Context Protocol paired with retrieval-augmented generation for schema context
• Data transfer and processing separated from reasoning, using AWS Batch for containerized downloads and nf-core bioinformatics pipelines
• Security implemented through Amazon Cognito authentication, AWS STS temporary credentials, AWS Secrets Manager, and IAM access controls
• Pattern generalizes beyond genomics to other disciplines with fragmented public data repositories like climate and public health research

VDOT demonstrates how conversational AI agents can make complex public data repositories more accessible, reducing technical barriers for researchers and supporting FAIR data principles.