Agentic antimicrobial resistance surveillance with AWS HealthOmics and Amazon Bedrock AgentCore

This article presents a reference architecture for agentic antimicrobial resistance (AMR) surveillance that transforms public sequencing data into plain-language answers about resistance genes without requiring bioinformatics coding skills.

• Ingestion stage pulls bacterial sequencing reads from the public NCBI SRA mirror using Step Functions and Fargate, with DynamoDB deduplication to avoid reprocessing.
• AWS HealthOmics runs a managed Nextflow workflow combining read quality control (fastp), de novo assembly (SKESA), and resistance gene detection (AMRFinderPlus).
• Lambda function normalizes detection output to hAMRonization schema; AWS Glue merges results idempotently into Apache Iceberg tables on Amazon S3 Tables.
• Strands agent on Amazon Bedrock AgentCore converts natural-language questions into Athena SQL queries against Iceberg tables with grounding via system prompts and Bedrock Guardrails.
• React web application with Amazon Cognito authentication and API Gateway provides user interface for submitting accessions and conversing with the agent.
• Entire solution defined in AWS CDK for single-command deployment with no hardcoded identifiers or manual setup steps.

The architecture eliminates manual bioinformatics workflows by automating public data ingestion, processing, harmonization, and enabling conversational exploration of AMR surveillance results.