Continuous Variant Reinterpretation Workflow
Built a workflow to automatically re-evaluate previously sequenced variants as new evidence emerges, surfacing cases whose classification has changed.
A clinical genomics group engaged Doppeldata to solve a problem that grows quietly over time: a variant classified today may warrant a different classification in a year as new evidence accumulates. They needed a way to reinterpret past results continuously - not re-sequence, just re-evaluate - and surface the cases that had genuinely changed. The catch is that the burden of this problem scales with success: the more genomes a lab has analysed, the larger the back-catalogue of variants sitting on classifications that were correct when made but may no longer be, and the less feasible it becomes to periodically revisit them all by hand.
- Built an automated re-evaluation workflow over previously classified variants
- Surfaced only the cases whose classification materially changed, not noise
- Turned a one-off sequencing result into a living, self-updating interpretation
The Challenge
A variant interpreted once is not interpreted forever - evidence databases and the literature move on, and a classification can shift from uncertain to pathogenic (or the reverse) without anyone re-examining the original case. A variant of uncertain significance is the clearest example: many are uncertain only because the evidence was thin at the time, and as population databases grow and case reports accumulate, a real fraction of them become classifiable. But nothing surfaces that on its own. The original report is filed, the classification sits frozen, and the new evidence that would change it lands in databases the report never looks at again. The group needed to re-evaluate a growing back-catalogue of variants automatically as the underlying evidence changed, and - just as importantly - to flag only the ones where the change actually mattered clinically, rather than drowning reviewers in trivial score fluctuations that alter nothing.
Key objectives:
- Continuously re-evaluate previously classified variants as the underlying evidence updates
- Surface only clinically meaningful classification changes, suppressing noise and cosmetic fluctuations
- Do it without re-sequencing - reinterpretation over existing data, not repeat lab work
The Approach
The workflow was designed to run continuously and stay signal-focused - the second property mattering as much as the first, since a re-evaluation system that cries wolf is quickly ignored:
Evidence tracking
Automated tracking was set up over the evidence sources feeding each variant's classification - the population-frequency data, clinical-significance assertions, and predictor outputs that the original call depended on - so that a change in any of them could be detected as it happened rather than at some scheduled manual review months later.
Re-evaluation engine
An engine re-ran the classification logic over the existing variant back-catalogue whenever relevant evidence changed, and compared each fresh result against the stored classification. Crucially this reused the same deterministic logic the variants were first classified under, so a change in the outcome could be attributed to a change in the evidence rather than a change in method - the difference had to be real, not an artefact of re-running things differently.
Change surfacing
The output was filtered to surface only materially changed classifications - a variant crossing a clinically meaningful boundary, such as VUS to likely pathogenic - while suppressing the trivial fluctuations that don't alter the call. This kept clinical attention on the handful of cases that genuinely warranted a second look, and kept the system credible enough to be worth watching.
The Results
- Delivered a workflow that re-evaluates past variant classifications automatically as evidence emerges, without anyone having to remember to revisit an old case
- Surfaced the small set of cases whose classification had materially changed - the ones crossing a clinically meaningful line - rather than a flood of cosmetic noise
- Converted static, one-off sequencing results into a continuously-updating interpretation, so the value of a genome kept pace with the science rather than freezing at the date it was reported
Why It Mattered
The value in a genome isn’t fixed at the moment it’s sequenced - it grows as science does, but only if someone reinterprets it. Automating continuous reinterpretation meant clinically important changes surfaced on their own, instead of sitting unnoticed in an old report.
Bioinformatics, statistical genomics & computational biology - practical tools, learning resources, and expert guidance based in South Australia. We are a community that believes in accessible knowledge and learning together.