Skip to content

DRAGEN pipelines

Most laboratories do not run a workflow repo — they run a vendor pipeline. Aver treats that as a first-class pipeline: the configuration is the pipeline. For Illumina DRAGEN, the snapshot comes from the replay JSON that DRAGEN writes with every run, so no instrument access is needed and nothing has to change in how you run it.

A DRAGEN snapshot carries:

  • the DRAGEN software version,
  • the reference hash-table digest and version,
  • the effective option set (sanitised — see below),
  • your licensed features and instrument software, declared in aver.yaml.

It is hashed and drift-classified exactly like a workflow snapshot: a DRAGEN upgrade or a reference hash-table change shows up as drift with rule-derived severity, and the evidence pack records the configuration your validation was run against.

Setup

apiVersion: aver/v1
pipeline:
  name: dragen-wgs
  engine: dragen
  # or pass --replay to aver resolve instead of setting it here
  replay: runs/latest/dragen-replay.json
  features: [cnv, hla]                     # your licensed DRAGEN features
  instrument:
    name: novaseq-x-control-software
    version: "1.3.1"
  clinical: true
  intended_use: Germline SNV and indel detection …
reference:
  build: GRCh38

The flow

# 1. Snapshot the configuration from a replay file.
aver resolve --replay runs/261001/dragen-replay.json -o snapshot.json

# 2. Run DRAGEN on GIAB reference reads yourself, then compare its VCF.
aver compare --query hg002.dragen.vcf.gz --sample HG002 \
  --snapshot snapshot.json --output aver-out/dragen-hg002

# 3. Upload the pair; drift now covers your DRAGEN configuration.
aver push --results aver-out

aver compare --snapshot stamps the snapshot hash into the result and writes a push-ready bundle, so the platform links the metrics to the exact configuration.

Configuration monitoring alone

Steps 2 and 3 need a benchmark run. To monitor configuration drift before the first benchmark — or between benchmarks — push the snapshot by itself:

aver push --snapshot snapshot.json

The platform records it and classifies drift against the validated baseline exactly as it would for a snapshot that arrived with results.

What leaves the machine

Replay files contain run-specific values that must stay in the laboratory, and Aver's scrub removes them client-side, before anything is sent:

  • options that carry sample identifiers by construction (read-group fields such as RGSM, sample and pedigree options, output-naming prefixes) are dropped entirely — two runs that differ only in sample naming produce the same snapshot;
  • path-valued options are kept by name with the value replaced by [path];
  • the raw command line is never stored.

aver resolve prints the scrub audit (how many options were dropped and redacted) every time, so what left the machine is visible.

Note

DRAGEN™ is a trademark of Illumina, Inc. Aver reads the output files your DRAGEN runs already produce; it is not affiliated with, endorsed or certified by Illumina.