Patient-context gene restriction filters

These are named gene filtering policies, separate from anatomical tissue sets and from the general adaptive CTA screen. No disposable-tissue exception is included. Both require positive evidence in their selected reproductive subset.

Machine name	Permitted positive normal expression	Mandatory exclusion
male_reproductive_maternal_fetal_excluded	Explicitly postnatal male reproductive contexts in male samples	All female/maternal tissues, placenta, other extraembryonic contexts, embryo and fetus, including fetal testis
female_developmental_male_reproductive_excluded	Female reproductive contexts in female samples; embryonic/fetal and extraembryonic contexts	Male reproductive anatomy at every stage and every other postnatal context; known male mammary expression is excluded

The first is the pregnancy-oriented candidate filter. The second is the male reproductive preservation candidate filter. Neither name asserts that a therapy is safe in pregnancy or preserves sperm/fertility. The second is deliberately conservative about fetal male reproductive anatomy: its explicit male-anatomy veto takes precedence over the larger fetal union.

The broad Reproductive set remains the union of all reproductive/developmental contexts. Testis plus oocyte can fit that set while failing both narrower profiles. Mammary tissue's anatomical grouping does not establish its sample sex.

Positive compatibility versus supported filtering

evaluate_normal_tissue_gene_filter() distinguishes:

• excluded: a measured positive lies outside the selected policy.
• insufficient_evidence: no counterexample establishes exclusion, but a positive or required negative context/metadata is missing or unusable.
• passes_declared_panel: an allowed positive exists and every supplied required negative context is assessed and negative, with the minimum domains represented.

The result always reports clinical_safety="not_assessed". It also records the filter policy version, taxonomy fingerprint and a SHA-256 fingerprint of the exact required coverage panel.

The caller fixes a named coverage panel independently of the gene. The pregnancy profile requires female somatic, female reproductive, pregnant maternal, embryonic, fetal, fetal male gonadal, fetal female gonadal, placental and other extraembryonic negative coverage. The male-preservation profile requires male reproductive, male somatic and male mammary coverage. These are minimum domain checks, not an exhaustive panel: the declared panel must specify relevant organs, routine cell types, ages/stages and modalities. One sample per domain cannot demonstrate biological absence throughout that domain.

Each negative is matched to the exact species, source/version, native tissue/cell labels, source stage, sex, physiological state, context/sample ID, assay and detection policy. Negative RNA cannot replace missing IHC; negative adult tissue cannot clear a fetal sample; one dataset cannot clear another. Every supplied positive overrides negatives, including positives outside the declared panel. Unassessed observations prevent passing.

No validated pregnancy coverage panel or corresponding retained gene list is bundled. The current adult HPA CTA report does not have the necessary maternal and fetal coverage and must not be presented as such a list. Fetal expression of MAGE-A1, GAGE and NY-ESO-1 has been observed in developing gonads; adult testis restriction does not resolve this gap. Primary fetal-gonad study

API

from oncoref.normal_tissues import resolve_normal_tissue
from oncoref.normal_tissue_gene_filters import (
    NormalTissueEvidence,
    NormalTissueCoverageRequirement,
    normal_tissue_gene_filter_profiles,
    evaluate_normal_tissue_gene_filter,
)

normal_tissue_gene_filter_profiles()
annotation = resolve_normal_tissue(
    "testis", species="human", stage="adult", sex="male",
)
positive = NormalTissueEvidence(
    gene_id="ENSEMBL:example",
    annotation=annotation,
    context_id="reviewed-source:sample-1",
    assay="RNA",
    detection_policy_id="reviewed-RNA-rule-v1",
    result="detected",
    evidence_id="reviewed-source:record-1",
)
decision = evaluate_normal_tissue_gene_filter(
    [positive],
    profile="male_reproductive_maternal_fetal_excluded",
)
assert decision["positive_pattern_compatible"]
assert decision["status"] == "insufficient_evidence"

NormalTissueCoverageRequirement uses the annotation, context ID, assay and detection-policy ID of each required negative context. Supply these as required_contexts with a coverage_panel_id. Assay-specific code supplies detected, not_detected or unassessed; absent rows, invalid values and unreliable IHC must never be converted to negative calls. This API does not silently reuse the general CTA screen's permissive RNA ceilings to assert absence.

Datasets and species

Anatomical definitions are shared; source mappings and expression evidence remain species-specific. Dog lacks the mapped seminal-vesicle/bulbourethral concepts; mouse coagulating gland is a prostate descendant. Uterine horns occur in the dog/mouse inventory. Mouse tonsil and dog/mouse vermiform appendix are inapplicable, not measured negatives. Unmapped head/foot/soft-tissue contexts cannot be narrowed by guessing. No mouse expression atlas is bundled.

The evaluator rejects mixed species, genes and taxonomy versions. Orthology does not transfer expression absence. Canine expression sample annotations now retain source sex where supplied; unknown source sex remains unspecified.