Automating enzyme design workflows with serverless, event-driven AWS

A Healthcare & Life Sciences business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had three focus areas.

For protein folding, a FASTA file uploaded to an Amazon S3 bucket raises an event that Amazon EventBridge routes to an AWS Lambda function. Lambda launches an Amazon ECS task on AWS Fargate, which reads the sequence from S3, runs the ESM model and writes the resulting PDB structure back to a designated S3 output bucket, all without anyone touching a container.

The delivered architecture automates both of a third-party provider's core modelling workflows end to end, removing manual container and instance management and adding resilience for its longest-running jobs. Acceptance is defined against S3-triggered automated execution of the containerised folding model, and batch processing with spot compute, automatic restart and checkpointing.

• Hands-off execution — A file upload now starts a full modelling run, with event-driven triggering replacing manual container and instance management.
• Cost-optimised compute — AWS Batch runs GPU jobs on spot instances first, falling back to on-demand only when necessary, to keep compute spend down.
• Resilient by design — Checkpoint-based restart automatically resumes long OpenAWSEM simulations after a spot interruption, protecting hours or days of work.
• Secure access — Task roles grant least-privilege access to Amazon S3 and AWS Secrets Manager, keeping credentials and data controlled.
• Handover — The pipelines are delivered into a third-party provider's AWS account with the architecture documented for the team to run and extend.

With its modelling pipelines running automatically, a third-party provider can spend less time managing infrastructure and more time designing the enzymes that make hard-to-recycle materials recyclable, ready to scale as its research grows.