Silent false normals: the error nobody sees

In clinical genomics the most dangerous error is not a wrong answer. It is a right-looking answer that was never actually computed. We call it the silent false normal: a variant a pipeline reports as absent, not because it was ruled out, but because it was never seen.

Where reads go to disappear

Standard short-read pipelines align against a single linear reference genome. In regions of high homology — pseudogenes, segmental duplications, repeat expansions — reads cannot be placed uniquely. Their mapping quality collapses to zero, the caller discards them, and the region returns empty.

A well-known example is CYP2D6, the pharmacogene that metabolises a large fraction of common drugs. In highly homologous regions such as CYP2D6/CYP2D7, conventional short-read alignment can produce large fractions of ambiguously mapped reads, impairing reliable copy-number and star-allele calling — so a poor-metaboliser phenotype can go unreported.

Genes that hide by design

The same failure mode affects SMN1/SMN2 (spinal muscular atrophy), PMS2 (Lynch syndrome) and GBA (Parkinson’s risk) — each shadowed by a near-identical pseudogene. A linear pipeline does not flag these as uncertain. It simply reports them as normal.

Declaring the region instead of omitting it

The first fix is not a better caller — it is honesty about coverage. If mapping quality and depth are recorded per region and reported alongside the findings, an unresolved region stops looking like a negative result. The geneticist can then decide: accept it, order an orthogonal assay, or state the limitation in the report.

Pseudogene-aware calling is the second step: modelling the paralogue explicitly so that signal can be separated from its shadow. At Innovare it is implemented for CYP2D6 through a specialised locus-specific analysis — described here.

The goal is not more variants for their own sake. It is that a report never presents an unexamined region as a confirmed normal.