AWS HealthOmics Streamlines AMR Gene Detection from Raw Reads

Bacterial antimicrobial resistance (AMR) poses a significant threat to global public health, with surveillance programs working tirelessly to monitor and mitigate its spread. The availability of public sequencing data, such as those from the NCBI Sequence Read Archive (SRA), provides valuable resources for this task. However, the challenge lies in transforming raw sequencing data into actionable insights, which traditionally requires significant bioinformatics expertise.

A recent post by AWS Architecture introduces a novel reference architecture that leverages AWS HealthOmics and Amazon Bedrock AgentCore to streamline this process. This architecture enables researchers to obtain plain-language answers about resistance genes without needing to write bioinformatics code or SQL queries.

What Changed with AWS HealthOmics?

The new architecture leverages AWS HealthOmics to automate the process of detecting resistance genes from raw sequencing reads. Traditionally, this would involve assembling a genome, screening it with specialized tools, loading the results into an analytical store, and writing SQL queries. Each step requires specialized knowledge, particularly in bioinformatics, making it cumbersome for epidemiologists who need these insights for public health decisions.

This solution begins with the ingestion of public sequencing data, facilitated by Amazon EventBridge Scheduler and AWS Step Functions, which automate the retrieval and conversion of sequencing runs from the SRA mirror on AWS. Following ingestion, AWS HealthOmics runs a managed Nextflow workflow for read quality control and resistance gene detection, eliminating the need for manual intervention and bioinformatics expertise.

How It Works

The architecture is composed of several stages:

• Ingestion: Utilizes AWS Step Functions and Amazon ECS on AWS Fargate to fetch and convert sequencing runs into a format suitable for analysis. This process is initialized by an Amazon EventBridge Scheduler rule or an on-demand REST call, ensuring flexibility in operation. The sequencing data is downloaded from the SRA mirror on the Registry of Open Data, and Amazon DynamoDB records processed isolates to prevent redundant processing.
• Genomics Processing: A Nextflow workflow on AWS HealthOmics chains together quality control with fastp, assembly with SKESA, and gene detection with AMRFinderPlus. The integration of these tools into a single workflow ensures a seamless transition from raw reads to assembled contigs and ultimately to gene detection, mirroring the typical community AMR pipeline structure. This workflow is particularly advantageous because it uses container images from Amazon Elastic Container Registry (Amazon ECR), with the AMRFinderPlus reference database embedded directly in the image to accommodate the lack of internet access during HealthOmics tasks.
• Harmonization and Storage: AWS Lambda normalizes detection output into the hAMRonization schema, and AWS Glue merges the results into Iceberg tables on Amazon S3. This step ensures that the data is structured and stored in a way that facilitates efficient querying and analysis. The idempotent nature of the AWS Glue job, which uses a MERGE operation keyed on the isolate and gene, ensures that rerunning the same isolate updates existing rows without creating duplicates.
• Analytical Store: Amazon Athena queries the harmonized data, providing a fast SQL surface for analysis. This stage leverages the AWS Glue Data Catalog to maintain a consistent schema across the Iceberg tables, enabling low-latency access to the data. By storing results in a structured format, the system supports complex queries that can be executed swiftly.
• Conversational Query Layer: A Strands agent on Amazon Bedrock AgentCore translates natural language queries into SQL, allowing users to explore data conversationally. This feature is supported by Amazon Bedrock Guardrails, ensuring that responses are grounded in the data. The web application interface, built with React and hosted on Amazon S3 and Amazon CloudFront, facilitates user interaction and is secured by AWS WAF and authenticated through Amazon Cognito.

Implementing AWS HealthOmics in Your Workflow

For those managing similar surveillance systems, integrating AWS HealthOmics can significantly reduce the complexity of genomic data processing. Here are some practical steps to consider:

• Set Up AWS Environment: Use AWS CDK to deploy the solution in your account. This deployment includes all necessary components, from data ingestion to the web application interface. The deployment is straightforward, requiring only a single command to initiate, and the accompanying code is available for customization.
• Automate Data Ingestion: Implement a workflow using Amazon EventBridge Scheduler and AWS Step Functions to automate the retrieval of sequencing data from public archives. This automation minimizes the need for manual intervention and ensures that the data pipeline remains efficient and reliable.
• Leverage Managed Workflows: Utilize AWS HealthOmics for running Nextflow workflows that handle quality control, assembly, and gene detection, reducing reliance on in-house bioinformatics expertise. The workflow is designed to run without internet access, with databases baked into container images, ensuring security and consistency. It is advantageous to pass container image URIs as workflow parameters, which maintains independence from specific AWS Regions and accounts.
• Ensure Data Harmonization: Use AWS Lambda and AWS Glue to normalize and merge detection results, ensuring that your data is consistent and ready for analysis. The use of the hAMRonization schema standardizes the output, making it easier to integrate and compare results across different datasets.
• Enable Natural Language Queries: Deploy a Strands agent on Amazon Bedrock AgentCore to allow users to query the data using natural language, enhancing accessibility and usability. This capability empowers non-technical users to interact with complex datasets intuitively.

Limitations and Considerations

While this architecture offers significant advantages for AMR surveillance, there are some limitations and considerations to keep in mind:

• Not a Clinical System: This architecture is intended as a demonstration and may require adaptations for specific clinical or regulatory environments. Users should assess the suitability of the system for their particular use cases, especially in environments with stringent compliance requirements.
• Cost Considerations: Running costs are tied to the volume of data processed and the frequency of queries. While inexpensive when idle, costs can rise with increased use. It is essential to monitor usage and optimize workflows to manage expenses effectively.
• Data Privacy: Ensure compliance with data privacy regulations, particularly if integrating with other datasets or sharing results across borders. This consideration is crucial for maintaining trust and adherence to legal requirements.
• Model Availability: The availability of models in Amazon Bedrock varies by AWS Region. Verify that the necessary models are available in your region before deployment. This step is critical to ensuring that all components of the architecture function as expected.

By leveraging AWS HealthOmics and Amazon Bedrock AgentCore, the architecture simplifies the path from raw sequencing data to actionable insights on antimicrobial resistance. This approach democratizes access to bioinformatics capabilities, allowing epidemiologists and public health professionals to focus on their core mission without needing deep technical expertise in genomics.

For those interested in exploring this architecture further, AWS provides comprehensive documentation and a GitHub repository to help you get started. The architecture demonstrates how cloud solutions can transform complex workflows into manageable, automated processes, paving the way for more accessible and efficient public health surveillance.