GRAIL Develops a Pioneering Multicancer Early Detection Test Using AWS

Launched in 2021, the Galleri test takes genetic data from a single blood draw and screens for a cancer signal by analyzing DNA methylation patterns. The team uses AWS to support the commercial scaling of the infrastructure to meet high demand and to fuel the software that runs its labs. The infrastructure uses over 60 AWS services.

For the compute resources to run Galleri tests at scale, GRAIL uses Amazon Elastic Compute Cloud (Amazon EC2), which provides secure and resizable compute capacity for virtually any workload. “One of the biggest values of using AWS is that we can concentrate up the stack without needing to worry about scale associated with storage or compute,” says Alag. To cost-efficiently run its computational workloads, the company uses Amazon EC2 Spot Instances, which let users take advantage of unused Amazon EC2 capacity. For its databases, GRAIL uses Reserved DB instances for Aurora, which provide a significant discount compared to On-Demand database instance pricing.

The GRAIL team developed Reflow to manage its bioinformatics workloads on AWS. Reflow language helps bioinformaticians to compose existing tools—packaged in Docker images—using ordinary programming constructs. Reflow runtime is deployed in Amazon Elastic Kubernetes Service (Amazon EKS) clusters, a managed service to run Kubernetes in the AWS cloud and on-premises data centers. It evaluates Reflow programs and parallelizes workloads onto Spot Instances, further reducing costs. It also improved performance through incremental data processing and memoization of results. “We are constantly looking for opportunities to optimize our architecture and to get the boost of using AWS services that we haven’t used before and changing our architecture to take advantage of those,” says Alag.

To address its storage needs, GRAIL uses Amazon Simple Storage Service (Amazon S3), an object storage service offering industry-leading scalability, data availability, security, and performance. The company has achieved cost savings using Amazon S3 Intelligent-Tiering (S3 Intelligent-Tiering), which automates storage cost savings by migrating data when access patterns change. “We transitioned most of our data to S3 Intelligent-Tiering, which led to 40 percent savings per gigabyte of storage cost,” says Ignatova.

Because GRAIL deals with sensitive health-related information, having a strong networking and security program is imperative. To make sure that its data is secure and complies with data privacy laws, GRAIL uses Amazon Virtual Private Cloud (Amazon VPC). It lets organizations define and launch AWS instances in a logically isolated virtual network, with guardrails in place to control access to sensitive data. “AWS provides really good infrastructure and capabilities that we use for data protection and encryption at rest and in transit,” says Alag. “We’re making use of the controls on AWS to restrict access to our sensitive data.” GRAIL expands into different AWS Regions and scales globally while meeting the data residency requirements by using the 87 Availability Zones on AWS.

In 2021 GRAIL partnered with the National Health Service (NHS) of England to implement Galleri in the largest multiyear, multicancer early detection trial to date, including 140,000 participants at mobile clinics operating in 150 locations around England. Those participating were recruited in a record 10 months. The enrollment ended in July 2022, and screenings are scheduled to continue for participants annually for 3 years. The NHS might eventually roll out the Galleri test to an additional one million people and has a long-term goal of detecting 75 percent of cancers while they are less advanced.