
AWS HealthOmics Explained for Beginners: Workflows, Sequence Stores, and Variant Stores
This article walks through what HealthOmics is, its main building blocks (sequence stores, workflows, and variant stores), and what has changed recently, so you don't build on a feature that is no longer open to new customers.
Series: AWS (6 articles)
- …
- 3AWS HealthOmics Explained for Beginners: Workflows, Sequence Stores, and Variant Stores This article
If you have ever run a genomics pipeline, you know the pattern. A few hundred gigabytes of sequencing reads arrive, someone writes a shell script, a server runs for days, and three months later nobody remembers which reference genome or tool version produced the results. AWS HealthOmics is a managed service built to take that pain away.
This article walks through what HealthOmics is, its main building blocks (sequence stores, workflows, and variant stores), and what has changed recently, so you don't build on a feature that is no longer open to new customers.
What is AWS HealthOmics?
AWS HealthOmics is a purpose-built service for storing, processing, and analyzing "omics" data: genomics, transcriptomics, proteomics, and related datasets. It is aimed at bioinformaticians, clinical researchers, and biotech and healthcare teams.
Think of it as three layers:
Layer | What it does | Main component |
|---|---|---|
Storage | Keeps raw and aligned sequence data, plus reference genomes | Sequence stores, reference stores |
Workflows | Runs your analysis pipelines at scale | Private workflows, Ready2Run workflows |
Analytics | Queries variant and annotation data | Variant stores, annotation stores |
You don't have to use all three. Many teams use only workflows and keep their data in Amazon S3.
Part 1: Sequence stores (where your data lives)
The problem they solve
Sequencing data is huge. A single human whole-genome sample can run to hundreds of gigabytes in FASTQ, BAM, or CRAM format. Storing it as plain files in S3 works, but you have to manage naming, metadata, lifecycle rules, and access yourself.
How they work
A sequence store is a managed home for sequence files. You import data from S3, and HealthOmics stores each file as a read set. A read set is more than a file. It has metadata attached, such as sample ID, subject ID, and the reference it was aligned to, so you can search and organize data by what it is rather than by file path.
Key ideas for beginners:
- Read sets are the unit of storage. They hold FASTQ, unaligned BAM, aligned BAM, or CRAM data.
- Reference stores hold your reference genomes (FASTA), such as GRCh38. Aligned data can point to the reference it was built against.
- Storage tiers: data you rarely touch can move to a cheaper archive tier, which matters when you keep thousands of genomes for years.
- Access control works through IAM, so you can control who can read which stores.
- Importing is done with import jobs that pull files from S3 into the store.
Why use one instead of plain S3?
Mostly for the metadata and organization. If you have ten samples, S3 is fine. If you have ten thousand, a store that understands samples, subjects, and references saves a lot of effort.
Part 2: Workflows (where the analysis runs)
This is the part most people start with.
What is a HealthOmics workflow?
A workflow is a pipeline definition, a recipe that says "take these inputs, run these tools in this order, and produce these outputs." HealthOmics runs it for you on managed compute, so you don't launch or manage servers.
It supports the three standard bioinformatics workflow languages:
- Nextflow
- WDL (Workflow Description Language)
- CWL (Common Workflow Language)
If your lab already has Nextflow or WDL pipelines, you generally don't need to rewrite them. You package the definition, register it, and run it.
Two flavors
- Private workflows. You bring your own pipeline, including custom or in-house ones. Your tools run as container images that you store in Amazon ECR.
- Ready2Run workflows. These are pre-built, AWS-managed pipelines for common tasks. HealthOmics also has a guide for running Illumina DRAGEN in software mode inside private workflows.
A simple mental model of a run
- You start a run, specifying the workflow, an IAM role, an output location in S3, and input parameters.
- HealthOmics provisions compute for each task and runs your containers.
- Logs and metrics go to Amazon CloudWatch.
- Results land in S3, and you pay for the compute and storage the run used.
Because each run is recorded with its workflow version and parameters, you also get much better reproducibility. In genomics this matters a lot, since it is the question reviewers and regulators ask first.
Recent features worth knowing
- Batch runs: you can submit up to 100,000 workflow runs in a single API request (added June 2026). That is very helpful for cohort-scale studies.
- Nextflow profiles: you can choose environment-specific configuration at runtime through the engineSettings parameter (added September 2026).
- Resource fallback ordering: an omicsResourceFallbackOrder directive was added for WDL and Nextflow workflows (October 2026).
Part 3: Variant stores and annotation stores (read this carefully)
What they were
After you align reads and call variants, you end up with VCF files. A variant store was a managed place to load those VCFs and query them with SQL-style tools (through services like Amazon Athena). An annotation store did the same for annotation data such as ClinVar or gnomAD-style resources, so you could join variants with annotations.
The important update
Many older tutorials still present variant stores as a core feature. They are no longer available to new customers. AWS documentation states that variant stores and annotation stores were closed to new customers starting November 7, 2025, while existing customers can continue using the service as normal.
If you are starting a new project today, don't plan around them.
What to use instead
AWS documents several migration paths:
- Reference ETL logic. HealthOmics provides the earlier variant-store ETL logic as a reference implementation, so you can keep control of how variants are transformed.
- Custom ETL. You can write custom AWS Glue ETL jobs, or use open-source Hail or Glow on Amazon EMR, to transform variant data.
- S3 table buckets. Store the results in Amazon S3 table buckets and keep using your existing AWS analytics services.
- AWS partners. AWS also points to partners that offer a variant and annotation store alternative.
A practical modern pattern is: sequence store or S3 → HealthOmics workflow → VCF in S3 → Glue/EMR ETL → S3 Tables → Athena or QuickSight.
How the pieces fit together
A typical end-to-end genomics project on AWS looks like this:
- Ingest: upload FASTQ files to S3 and import them into a sequence store (or keep them in S3).
- Process: run an alignment and variant-calling workflow (Nextflow, WDL, or CWL) on HealthOmics.
- Store results: write VCFs and reports to S3.
- Analyze: transform and query variants with Glue, S3 Tables, and Athena.
- Share and visualize: use QuickSight or your own notebooks.
When should you choose HealthOmics?
It is a good fit when:
- You already have Nextflow, WDL, or CWL pipelines
- You process many samples and need scale
- Reproducibility and auditability matter, as in clinical or regulated settings
- You don't want to manage clusters or batch queues
Consider something else when:
- You run a handful of small jobs, where plain EC2 or AWS Batch may be simpler and cheaper
- Your tool isn't containerized and you can't easily containerize it
- You only need ad hoc docking or simulation, which isn't a standard omics pipeline
Cost basics (without the surprises)
Exact prices change by region and over time, so check the official pricing page. The cost drivers are:
- Workflow compute: billed for the resources each task uses while it runs, plus run storage
- Storage: sequence store data, with cheaper archive tiers for cold data
- Ready2Run workflows: usually priced per run
- Everything around it: S3, ECR, CloudWatch, and any analytics services you add
Two habits save money: right-size your task resources (over-provisioning is the most common waste) and archive data you don't touch.
Getting started: a beginner roadmap
- Run a Ready2Run workflow on a small public dataset to see the full flow.
- Create a sequence store and import a few files to understand read sets.
- Take a small Nextflow or WDL pipeline, containerize its tools, push the images to ECR, and register it as a private workflow.
- Add cost tags and review run metrics to tune resources.
- Build the downstream S3 Tables + Athena analytics layer instead of relying on variant stores.
Key takeaways
- HealthOmics = storage + workflows + analytics for omics data.
- Sequence stores organize large sequence files with sample-level metadata.
- Workflows run Nextflow, WDL, and CWL pipelines without server management, and recent additions like batch runs make cohort-scale work easier.
- Variant and annotation stores are closed to new customers, so use S3 Tables with Glue or EMR for new variant analytics.
- Choose HealthOmics for scale and reproducibility, and simpler compute for small one-off jobs.
Features and availability change quickly. Check the official AWS HealthOmics documentation before designing a production system.
Series: AWS (6 articles)
- …
- 3AWS HealthOmics Explained for Beginners: Workflows, Sequence Stores, and Variant Stores This article
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article