How to validate a whole-genome pipeline for accreditation
A practical walk through truth sets, acceptance criteria and the paperwork inspectors actually ask for.

Validating a whole-genome pipeline is less about any single benchmark and more about showing, clearly and repeatably, that the pipeline does what you say it does.
Start with the intended use
Write down what the pipeline is for before you run anything. A germline WGS pipeline for rare disease has different acceptance criteria from one used for research cohorts. Your intended use decides which variant types matter, which regions you report on and how much sensitivity you need.
Choose truth sets that match your samples
Public reference samples such as the Genome in a Bottle set are the starting point. Add in-house samples with orthogonally confirmed variants, especially for the variant types and genes you report most often.
- Use at least one sample per ancestry group you commonly test
- Include known difficult regions, not just the easy ones
- Record the exact version of every truth set
Set acceptance criteria up front
Decide on sensitivity and precision thresholds for SNVs, indels and CNVs before you look at the results. Changing thresholds after the fact is the fastest way to lose an inspector's trust.
Lock the versions
Every tool, reference genome and parameter should be pinned and recorded. When something changes, it goes through change control and a targeted revalidation, not a quiet update.
Keep the evidence together
The validation report, the raw benchmark outputs and the version manifest should live in one place. On Strand, the validation runs, their results and the locked pipeline version are linked, so the evidence is one click away when an inspector asks.


