Linked tissue, cell-cluster and immune RNA thresholds
This research comparison uses one tissue/IHC ladder for bulk RNA, HPA cell clusters and sorted immune RNA. Source-specific numerical cutoffs are not independently tuned. The original reproductive/developmental association, normal-protein exclusion, RNA coverage and measurement-support requirements remain. This does not change the released expression classes or forbidden sequence reference.
Results
| Tumor admission | Shared normal-RNA ladder, including corneal count scaling | Sensitivity retaining the old 1 CP10K corneal cutoff |
|---|---|---|
| ≥1 pTPM in ≥3 samples and ≥5% of one cohort | 244 | 248 |
| ≥10 pTPM in any one measured sample | 337 | 341 |
| ≥30 pTPM in any one measured sample | 197 | 201 |
The ≥10 result gains 99 and loses six compared with the recurrence result; it gains 140 compared with the ≥30 result. Relative to the old 201-gene screen, the linked recurrence result gains 46 and loses three. These comparisons keep their tumor rules explicit rather than attributing every difference to normal RNA thresholds. All 20,097 genes were reprocessed; 2,960 qualified for detailed local assessment, producing 793,280 cell-context records.
SMC1B and PAGE1 pass all three linked tumor scenarios. SMC1B's largest DICE third-highest value is 3.4997 TPM, below the shared Protected ceiling of 6 TPM. Its stimulated-CD4 population has 58/88 reported values ≥2 TPM, but none ≥5; the measured finding is retained even though the arbitrary 2 TPM veto is gone.
The eight requested identities MAGEA4, CTAG1A, CTAG1B, XAGE1A, XAGE1B, SSX2, CT83 and TBXT pass. PRAME and CTAG2 still fail independent Vital tissue RNA and bulk gates. There are no named-gene overrides.
The corneal sensitivity differs by DKKL1 (0.041542 CP10K), INSL4 (0.112708), DCAF4L1 (0.021790) and RBAK-RBAKDN (0.078576). These are supported recurrent statistics, not individual-cell maxima.
The 337 carry 141 historical CTA, 9 Unsafe and 187 Neither annotations. Those are existing expression-policy labels, not new classifications. The nine historical Unsafe flags are retained for TNP1, OBP2A, SMCP, XAGE3, HMGB4, FTMT, TMEM95, LELP1 and BOD1L2. Their presence must not be read as removal of a previous protein/antibody concern or as clinical clearance.
Selected 337-gene table with retained caveats.
Where the conflicting numbers came from
The 21 October registry defined inclusive candidate ceilings of Vital 2/4/6/12, Protected 6/8/12/24, and Manageable 10/20/30/60, ordered by Missing-or-Uncertain, Approved, Supported and Enhanced negative IHC. Those are user-directed project parameters, not clinical ground truth.
The 26 October illustrative cell screen independently introduced lower/upper pairs of 2/5, 5/10, and 10/20, and classified all sorted immune assays as Vital. That is why DICE acquired a 2 TPM candidate exclusion. Neither DICE data nor a validated single-cell purity model established this reduction. The old 201-gene comparison inherited that separate grid.
The protected 6 and 12 in the earlier registry are its Missing and Supported IHC ceilings, respectively. They were not originally two validated candidate and toxicity boundaries. In this revision, stronger RNA reference review starts at 2× the candidate ceiling, uniformly across tiers and assays. Thus a Protected context without usable IHC now has a candidate ceiling of 6 and a stronger-review boundary of 12. That factor of two is an explicit provisional policy choice, not evidence of protein production or toxicity.
One shared ladder
| Tissue or cell tier | Missing/unusable IHC | Approved negative | Supported negative | Enhanced negative |
|---|---|---|---|---|
| Vital | 2 | 4 | 6 | 12 |
| Protected | 6 | 8 | 12 | 24 |
| Manageable | 10 | 20 | 30 | 60 |
Values at the candidate ceiling pass. Values above it exclude from candidate selection; values at or above twice the ceiling additionally enter stronger RNA reference review. RNA review does not automatically add a sequence to the forbidden reference. Reliable positive protein evidence remains separate.
Bulk and cell code both obtain candidate values from
normal_tissue_screening_policy("ihc_adaptive_cta")["negative_ihc_rna_limits"].
cell_rna_policy.linked_thresholds() generates every cell-assay row from that
definition. The old independent bulk cap of 10, cluster cap of 30, immune cap
of 5, cell IHC multipliers and corneal cap of 1 are not additional gates in the
linked result. Their presence as historical columns does not affect selection.
HPA bulk and cell data retain nTPM, DICE and Monaco retain TPM. Equal numerical settings are a provisional policy linkage, not a claim of calibrated assay equivalence. A source-specific deviation needs ground-truth calibration and a versioned justification; none are configured in this policy.
Negative IHC only changes a ceiling when there is a reliable reproductive-positive profile and no recorded antibody concern. Bulk relaxation requires all expected IHC contexts in the same anatomical tissue to be negative. A cell population requires the exact tissue and cell label. Missing, partial, unmatched or Uncertain IHC receives the baseline. Detected nonpermitted IHC, including Uncertain staining, still excludes selection.
Immune cells and reproductive context
Identified immune cells default to Protected, including DICE, HPA/Monaco sorted populations and mapped immune populations in HPA tissue clusters. Their nonimmune neighbors retain the organ-specific tier. No immune subset has a reviewed Manageable exception in this run; an anatomical location or presumed replenishment alone does not create one. This is a selection-policy overlay; the anatomical registry and measured values are preserved.
Explicit reproductive/placental/embryonic/fetal immune contexts stay neutral for exclusion and do not supply positive association. All 426 source HPA clusters still contribute to coverage: 399 screened and 27 newly neutral. The policy does not exempt normal childhood expression.
DICE still uses the third-highest reported value per population, so the candidate ceiling is exceeded only with at least three reported values above 6 TPM. HPA/Monaco use their source summaries. These are not invented independent donors or individual-donor maximum gates. No equivalent IHC match is assumed for sorted or activated immune states.
Corneal counts and the meaning of pTPM
pTPM means protein-coding transcripts per million, an RNA normalization. HPA removes noncoding transcripts and rescales the remaining abundance; nTPM adds its between-sample normalization. HPA methods.
Our adult Collin corneal statistic is the mean of per-cell counts divided by whole-cell library size and multiplied by 10,000, including zeros. Therefore CPM = 100 × CP10K exactly for that same statistic. With this 10x 3′ UMI assay, a bulk-fragment gene-length correction is not the way to convert it. Primary source.
The result is still not HPA-equivalent nTPM: protein-coding denominators, TMM, gene inclusion and pooled-cluster versus mean-per-cell aggregation differ. The linked policy uses the shared nominal per-million ceiling divided by 100 in native CP10K, without relabeling observations. For Vital corneal populations without IHC, 2 per million corresponds to 0.02 CP10K, with stronger review at 0.04. The original 1 CP10K = 100 CPM rule is exported as a separate sensitivity so this substantial numerical assumption is visible.
Corneal support remains ≥5 positive cells and ≥20 counts in each of ≥2 distinct sample groups. HPA tissue clusters still need ≥20 gene reads. Coverage is reported separately; low support and missing data are not interchangeable with a measured zero.
Tumor admission and reproduction
The main alternative asks for at least one measured tumor sample ≥10 pTPM in HPA 25.1, without a minimum prevalence or three-sample requirement. The same normal-expression rules are compared against the old recurrence rule and the one-sample ≥30 pTPM rule. These use 21 TCGA and ten validation cohorts, not all OncoRef cancer datasets. They are tumor nomination thresholds, not an inversion of a validated normal-tissue safety limit.
All 20,097 reference genes are reprocessed for context neutrality. The complete union passing association, protein, coverage and either tumor rule receives local bulk/cell assessment. Old RNA exclusions do not restrict that union.
python scripts/cta_quantitative_threshold_report.py \
--neutral-reproductive-immune --linked-tiers --tumor-floor 10 \
--output /path/to/empty-directory
Every gene and selection scenario, cell observations and decisions, bulk observations and decisions, generated shared cell thresholds, machine-readable counts, input/source/output hashes.
These remain adult-scope research parameters. No ground-truth toxicity, peptide-presentation calibration or pediatric normal-panel validation is claimed by a passing candidate or passing software tests.