Continuous normal-expression breadth

Descriptive policy oncoref.normal-expression-breadth.2026-10-26.1; active eligibility remains oncoref.target-sets.2026-10-25.2. All 2,540 mapped nominees and three prespecified ordinary-normal controls are represented. Scores do not admit or exclude any gene. Clinical benchmark genes were chosen before scoring; this is a descriptive comparison, not validation.

For each category, B = sum(weight * RNA / (RNA + ceiling)). Vital, Protected and Manageable are separate. Reproductive abundance never enters this formula. The complete profiles use the current archived per-gene IHC-adjusted RNA ceilings. A fixed-baseline comparison uses 2/6/10 nTPM for every gene, including ACTB, GAPDH and PPIA controls. All scores and weighting sensitivity · Benchmarks.

Fixed anatomy and missing coverage

The panel is specified independently of each gene's observed tissues. Each organ has equal weight within its category; eight brain regions divide one brain weight. Retina and unmeasured non-retinal eye divide one eye weight. Broad small intestine is used once, without adding overlapping duodenum. Bronchus and embedded vascular/immune populations are not counted as additional independent whole organs. Cellular blind spots still require the separate cell-expression audit. This panel is not whole-body cell coverage.

Blood, peripheral nerve, inner ear, non-retinal eye and several protected compartments are explicit missing coverage where this bulk source has no sample. Observed weighted contribution is the lower bound; adding missing weight gives the upper bound. These are coverage bounds, not confidence intervals. An all-missing gene gets [0,1], never a negative finding. Six nominees absent from the archived tissue screen retain their missing profiles in that scenario. Sensitivity halves/doubles the prespecified brain organ weight, conserving within-brain weights and renormalizing the fixed panel, never just observed rows.

Donors, sources and units

HPA consensus is a published cross-source summary, not a donor median or an independent cohort. Individual-donor counts and sampling intervals cannot be inferred from it. Available reviewed donor summaries currently cover CTAG2 heart only: four independent donors permit a median; the prespecified ten-donor minimum leaves q90 unavailable. Separate donor burdens and profiles use the fixed baseline ceilings, because these RNA donors are not paired to IHC. The unavailable q90 retains entirely missing coverage, never a zero measurement. A marginal 95% percentile-bootstrap interval uses 2,000 donor resamples with seed 0; it is reported separately from anatomical coverage and is unstable with four donors. This tiny selected sample is not an estimate of population risk, nor matched RNA/IHC specimens. No whole-profile sampling interval is claimed. Native TPM/CP10K, cell atlases, other species and HPA versions are excluded pending their own calibration.

Interpretation and reproduction

No classification cutoff has been fitted. Accepted focal protein findings, hard RNA exclusions and current holds remain in the eligibility policy; a low burden cannot erase them. This score is a review-priority descriptor, not injury probability or a body-wide expression fraction.

Run .venv/bin/python scripts/build_normal_expression_breadth.py --out <empty-directory>. The manifest pins inputs, implementation, source receipt and all outputs.