Filtering & prioritization

Interpretation starts from a shortlist, not from millions of raw calls. Two layers of filtering produce that shortlist.

Workflow-level filtering (already applied)

Before a case reaches the Interpreter, the analysis workflow has already:

• Removed common variants by applying a population-frequency cap. Variants more common than the cap in reference populations (e.g. gnomAD) are filtered out, on the principle that a variant common in the general population is unlikely to cause a rare disease.

• Annotated the remaining variants with gene, consequence, frequency, and clinical significance.

Case-level filtering (you control)

Within the Interpreter, each variant tab has its own filter drawer. Every variant list shares a User Classification filter (Pathogenic, Likely Pathogenic, VUS, Likely Benign, Benign); the rest vary by variant type:

• Small variants: User Classification, Gene, Chromosome, Position.

• Structural variants: User Classification, Automated Classification, Gene, SV Type (DEL, INS, DUP, BND, INV), Chromosome, Position.

• Copy number: User Classification, Type (DEL, DUP), Gene, Chromosome, Copy Number, Position.

• Tandem repeats: User Classification, Chromosome, Position, LPS Outlier.

• Paralogs: no filters.

Two tabs open with a filter already applied: Structural Variants is pre-filtered to Automated Classification = Pathogenic and Likely Pathogenic, and Tandem Repeats is pre-filtered to LPS Outlier. Clear or adjust these if you want to see the full list.

How to use filters well

1. Start from the phenotype: apply a relevant gene panel or HPO-driven prioritization if available.

2. Tighten on rarity and predicted impact to surface the strongest candidates.

3. Loosen deliberately if the strong candidates don't explain the phenotype. Keep a record of what you relaxed.

A frequency cap that is too strict can hide a real but slightly-more-common variant. If a case is unsolved, revisit the frequency threshold before concluding.