Thresholds, patient identity and named CTA placements
Corrected draft: 638 CTA, 260 Cancer Antigen only, 22 Reproductive Antigen only, 714 Other and 18,259 Unsafe. These use the existing fivefold single-tumor rule. Every protein · Named CTA families and studied targets · All threshold values as CSV.
Samples and patients
OncoRef's cancer matrices have one column per sample/library. The previous comparison resolved only TCGA, TARGET and MMRF patient barcodes, omitting patient metadata already available in OncoRef. expression.sample_metadata now joins actual matrix columns to the central samples.sample_identities policy. The rebuild and threshold analysis use that same mapping. Source-specific patient IDs, physical sample keys, status, basis and references stay separate.
The correction resolves 2,471 additional samples. 13,164 of 14,440 sample columns have documented identities, representing 13,034 patient IDs; 1,276 samples remain unresolved. Identities use each source's documented patient namespace; unavailable cross-study linkages remain unknown. A copied sample label in a GEO case field is not accepted as a patient mapping.
| Source namespace | Additional resolved samples |
|---|---|
| BEATAML_OHSU_2022 | 600 |
| CGCI_BLGSP | 182 |
| CLLMAP_2022 | 708 |
| OPENPBTA_V23_CBTN_CRANIO | 29 |
| OPENPBTA_V23_DIPG_H3K27 | 32 |
| TREEHOUSE_POLYA_25_01 | 905 |
| TREEHOUSE_RIBOD_25_01 | 15 |
The CLL-map source specifies six duplicate exclusions plus one suspected MCL exclusion, already applied to OncoRef's 708-sample cohort. Treehouse uses curated donor IDs where present and the existing builder's donor-family rule for older identifiers. Multiple S01/S02 libraries from one family count once. Complete source coverage.
Multiple samples are not mandatory. One qualified tumor ≥5 × max(N,1) still qualifies. Patient identity is needed only to establish independent support for the repeated route and patient-based percentiles. Unknown IDs do not erase measured expression.
| Protein | Before correction | Corrected category | Patients above normal |
|---|---|---|---|
| DEFB106A/B | Reproductive Antigen only | CTA | 2 |
| TKTL2 | Reproductive Antigen only | CTA | 34 |
| ARL13A | Reproductive Antigen only | CTA | 14 |
| OR4X1 | Other | Cancer Antigen only | 2 |
| OR56B2 | Other | Cancer Antigen only | 10 |
| OR4C15 | Other | Cancer Antigen only | 2 |
| OR4A15 | Other | Cancer Antigen only | 3 |
| KRTAP19-2 | Other | Cancer Antigen only | 4 |
| MINDY4B | Reproductive Antigen only | CTA | 2 |
| LACTBL1 | Other | Cancer Antigen only | 3 |
| FAM90A26 | Other | Cancer Antigen only | 2 |
| OR4F3 | Other | Cancer Antigen only | 3 |
| OR4C13 | Other | Cancer Antigen only | 4 |
| DPEP2NB | Reproductive Antigen only | CTA | 3 |
| PRAMEF13 | Reproductive Antigen only | CTA | 2 |
| FAM246B | Other | Cancer Antigen only | 2 |
The expression values and normal exclusions did not change. Current-rule changes and tenfold-alternative changes retain complete evidence. The baseline is commit 2e24b126; the threshold comparison has been regenerated with the corrected identities.
Threshold inventory
These are the rules used in the current draft or inherited normal screen. Proposed alternatives are labeled separately. Historical 1.36.0 used a ≥5 reproductive RNA threshold, tumor presence >0 and the >10% concentration veto; those are not the current draft's association rules. Older compatibility taxonomy parameters do not replace the active cta_selection_policy ceilings.
| Rule | Threshold | Status | Meaning |
|---|---|---|---|
| Cancer: repeated | ≥2 documented patients; each tumor ≥1 TPM and strictly above N | Current draft | N is the largest supported nonreproductive, non-Disposable normal context statistic; repeated libraries count once. |
| Cancer: single tumor | ≥5 × max(N, 1) TPM in ≥1 sample | Current draft | One strong sample suffices, including one without a resolved patient ID. |
| Reproductive bulk RNA | ≥1 native nTPM | Current draft | Includes reproductive, placental and developmental scope; normal exclusions still apply. |
| Reproductive-cell RNA | ≥1 nTPM and ≥20 reads | Current draft | Indirect sites also require complete outside-cluster coverage, every outside value <1, and no curated normal-induction conflict. |
| Developmental RNA | Mean ≥1 TPM; ≥20 cells, ≥5 positive and ≥10% positive; positive means ≥1 TPM | Current draft | Qualifying developmental lineage/stage; bulk fetal/reproductive scope is retained separately. |
| Reproductive protein | Reliable mapped positive IHC or reviewed human reproductive protein for every member | Current draft | An alternative to RNA association; antibody/identity conflicts remain unresolved. |
| Normal RNA ceilings | Vital 4/8/16/32; Protected 8/16/32/64; Manageable 16/32/64/128 | Retained normal screen | Columns are missing/uncertain, Approved, Supported, Enhanced matched negative IHC. Units remain native TPM or nTPM. |
| Reviewed clinical RNA credit | One IHC rank increment, capped at rank 3 | Retained normal screen | Only qualifying product-specific evidence; no stacking or override of injury/protein exclusions. |
| Normal RNA consequence | Above Vital/Protected ceiling → Unsafe; above Manageable ceiling → hold; Manageable >128 → Unsafe | Retained normal screen | Equality to the ceiling passes. Disposable and reproductive scope have no current numerical ceiling in this gate. |
| HPA tissue-cell support | ≥20 reads | Retained normal screen | Cell-specific tier; only exact/declared compartment-matched IHC can supply credit. |
| HPA/Monaco immune | Cell-specific 4/8/16 bases on native nTPM/TPM | Retained normal screen | No matched negative-IHC credit; a reproductive location does not exempt shared immune cells. |
| DICE immune | Third-highest reported TPM; ≥3 observations | Retained normal screen | Apply cell-specific ceiling; unpaired distributions keep their bounds. |
| Adult cornea | ≥5 positive cells and ≥20 counts in ≥2 sample groups; RNA ceilings divided by 100 | Retained normal screen | Native CP10K screening scale, not a TPM conversion. |
| Rare-cell detection | ≥1% detected, ≥3,000 cells and ≥10 expressing cells | Retained normal screen | Existing brain/spinal cord/heart/lung panel: Vital → Unsafe; Protected → review hold. |
| Normal protein | Reliable non-Low positive in Vital/Protected → Unsafe; Manageable → hold | Retained normal screen | Low is flagged; uncertain is unresolved. Reviewed/external findings retain reagent, cell and identity scope. |
| Required RNA coverage | Every nonreproductive combined-bulk tissue concept for every member | Retained normal screen | The comparison uses 36 concepts. Missing optional cell/immune/cornea panels alone does not block; observed exclusions still apply. |
| Translation entry | Reference coding CDS, or supported ncORF: MS peptide or ≥2 independent Ribo-seq studies | Current draft | All proteins still undergo the normal screen. |
| Representative translation | Complete primary-locus coding sequence; no LoF/NMD, incomplete tags, internal stops or non-AA20 residues | Current draft | MANE Select first, Ensembl canonical second; conflicting evidence requires review. X is not allowed. |
| Nested coding sequence | <50 amino acids inside a same-strand coding gene without independent ncORF evidence | Current draft | Other:Sequence. |
| Unsafe sequence overlap | ≥95% of 9-mers occur in Unsafe proteins | Current draft | Other:Sequence; this does not create expression-based Unsafe evidence. |
| Reproductive concentration | Recorded normal/(normal + reproductive maximum); previous >10% veto removed | Descriptive only | Developmental and adult reproductive association now share the same admission flow. |
| Published nomination | Exact-member claim in a curated CA/CTA/oncofetal target source | Annotation | Retain source, table and claim; neither automatic admission nor exemption from normal findings. |
| Proposed single-tumor rule | ≥10 × max(N, 1) TPM | Counterfactual; not adopted | Repeated route remains ≥2 patients at ≥1 TPM and above N. |
| Proposed target maximum | max(cancer, reproductive/developmental) ≥1 and strictly above every measured nonreproductive value | Counterfactual; not adopted | Source-compatible values only; missing compartments are not zeros. |
| Proposed Disposable caps | ≤32 TPM and ≤32 nTPM, assessed separately | Counterfactual; not adopted | No unit conversion or IHC-derived numerical allowance for these proposed caps. |
| Percentile alternatives | Within-tumor p80/p90/p95; across-patient p80/p90/p95 with ≥20 measured patients | Sensitivity analysis | Within-tumor ranks use the full measured catalog; across-patient values use the lower observed order statistic. |
The base normal ceilings are 4 for Vital, 8 for Protected and 16 for Manageable, before qualifying IHC or clinical adjustments. They are screening parameters, not measured injury thresholds. Normal-screen implementation, antigen membership implementation and the tissue registry with cited rationales identify the distinct policy layers.
Hair follicles
Hair follicles are already classified Disposable; whole skin is Protected. The missing component in the numerical cap comparison is isolated hair-follicle RNA, not a missing ontology bucket. HPA v23's tissue-cell RNA file has no hair-follicle cluster. It does have cutaneous melanocytes on its separate nTPM scale; the cell audit retains read support and missing loci.
HPA v23 also contains hair protein staining: cortex/medulla, cuticle, internal/external root sheath, and hair-follicle cells in skin sections. Thus “no hair data” was too broad. A separate mapping defect was found: the skin cell-label regex hair follic does not match hair follicles and falls back to Protected skin. Issue #687 tracks correction together with a replay of the inherited normal-protein gates and affected sequence containment; the patient-only correction here does not revise those gates.
Human hair-follicle RNA datasets are available, including GSE129611 and the human scalp single-cell study. They have not been added to the pinned cap panel. Missing hair RNA cannot count as zero Disposable expression.