Initial CTA literature audit — 25 October 2026
Historical review snapshot: the proposals below preceded the completed adjudications and current 213-gene policy. In particular, the PLCZ1 pancreatic protein interpretation was corrected to qualitative RNA; this frozen search table is not the current classification API.
I would not endorse all 399 retained genes as established safe targets. The current categories reproduce the approved screen, but newly examined literature supports five proposed exclusions and three review holds. These are proposed research classifications; the released sets have not been changed.
Search every gene · All-gene table · Proposed changes · Paper leads · Exact searches · Reproduction and limitations · General category proposal · Filed issues
What the current decisions mean
| Current result | Genes | Interpretation |
|---|---|---|
| CTA | 399 | Candidate passes the recorded screen; not established safe for every peptide, receptor, patient or tissue state |
| Neither | 561 | Neither a target nor default forbidden background; evidence is low, missing, variable or unresolved |
| Unsafe | 1,580 | Excluded under a normal-expression or safety rule; many have no demonstrated clinical toxicity |
| Total | 2,540 | Mapped source nominations, not all human genes or all historical CTA labels |
All 2,540 frozen category/detail assignments matched the 24 October API. No current CTA or Neither row has an audited non-permitted normal maximum above 30 TPM or nTPM. These are consistency checks, not independent validation of the biological cutoffs. The two units are checked separately; corneal CP10K is not converted to TPM.
The per-gene table explicitly distinguishes a policy-consistent decision from
completed primary-paper adjudication. All 2,540 genes were searched in 6,013
queries; 32 genes received new primary-paper assessments and 46 have specific
assessment notes including carried-forward reviews. The remaining 2,508 genes
have not received a new individual primary-paper adjudication in this pass.
There are 36,964 unique retrieved paper leads, 1,688 truncated searches and 160
genes with no hits in either topic query. No-hit searches do not establish safety.
Searching every gene does not mean every paper has been read or every result is
specific to that gene. summary.json records these coverage limits; unresolved
rows remain visibly provisional.
Changes I would propose
| Gene | Current → proposed | Evidence and interpretation |
|---|---|---|
| FTMT | CTA → Unsafe | Human substantia nigra protein in neurologically unaffected postmortem donors, with a monoclonal antibody, absorption controls and an H-ferritin specificity check. This was absent from the previous 37-gene conflict list. Primary study |
| ABCC11 | CTA → Unsafe | Human brain immunoblotting and axonal immunofluorescence conflict with reproductive restriction. Abstract reviewed; full reagent assessment remains useful. Primary study |
| YBX2 | CTA → Unsafe | The original dbpC/contrin study reports vascular smooth-muscle and adrenal-cortical protein. Specimen context and polyclonal-reagent limitations are recorded; vascular staining is not cardiomyocyte staining. Primary study |
| CCDC158 | CTA → Unsafe | Published renal tubular protein and endocytosis biology contradict unrestricted gene-level inclusion. A testis-specific isoform would require a separate sequence-level decision. Primary study |
| ACTRT1 | CTA → Unsafe | Endogenous protein in differentiating normal human epidermal keratinocytes, supported by multiple reagents and protein-band analysis. This is broader than permitted hair follicles. Primary study |
| MAGEC2 | CTA → Neither | Historical Purkinje-cell staining conflicts with the curated normal-negative profile. Resolve polyclonal and paralog specificity before unrestricted use. Primary study |
| LGALS13 | CTA → Neither | Extra-placental antibody reactivity conflicts with placental RNA restriction and later galectin work. Resolve related-galectin specificity and adult versus fetal compartments. Early study, later study |
| PLCZ1 | CTA → Neither | Historical NYD-SP27 work reports endogenous human pancreatic-duct expression. Full-paper and isoform/reagent adjudication remain necessary. Another omission from the previous conflict list. Primary study |
If all eight proposals were accepted, the mapped totals would become 391 CTA, 564 Neither and 1,585 Unsafe. The public API, FASTA files and k-mer backgrounds still use the released categories. None of these expression studies alone shows that a particular therapy injured the corresponding tissue.
Important findings that should not become automatic exclusions
| Finding | Why it matters | Appropriate response |
|---|---|---|
| 180 retained gene assignments / 1,707 primary-antibody normal-staining rows | Curated IHC can be negative while individual primary records stain non-permitted compartments | Review specificity and the curator's rationale; do not treat all raw stains as accepted protein |
| 103 retained assignments have Medium/High primary staining in Vital compartments | The discrepancy includes consequential cell types | Prioritize orthogonal evidence and reagent identity |
| 78 of the 180 have explicit off-target/cross-reactivity warnings | Some positive stains were deliberately discounted | A blanket raw-IHC veto would discard legitimate targets and duplicate family-level observations |
| PATE2, TPTE2, EQTN | Historical normal-tissue expression is isoform-dependent | Resolve transcripts/proteins before gene-wide or k-mer claims |
| ACTL7A and SPAM1 | Additional historical non-reproductive RNA not highlighted in the 37-gene list | Quantify and resolve identity; qualitative RT-PCR is not a current nTPM value or proof of protein |
| SUN5, SLC22A14, FAM170A, HORMAD2 and other recorded conflicts | Older tissue studies and modern measurements differ | Preserve both observations and distinguish species, RNA/protein and assay specificity |
| PASD1 | RNA in histologically normal tissue from cancer patients | Relevant nonmalignant context, but not healthy-volunteer protein evidence |
The antibody counts are gene assignments, not independent proteins, donors or clinical injuries. Shared family reagents can recur across loci. In particular, SAGE1 profiles explicitly describe presumed off-target binding being disregarded. An “Enhanced” antibody/profile label does not make every raw tissue stain gene-specific, nor does it prove all unmeasured tissues are negative.
Excluded genes worth revisiting
| Gene(s) | Current result | Assessment |
|---|---|---|
| ROPN1 | Unsafe / High RNA | Salivary serous-cell RNA is 32.9 nTPM, close to the hard cutoff. Review source identity, ambient RNA and protein before any rescue. |
| TDRD1, BRDT | Neither / Cellular | Modest immune-RNA signals near review thresholds; worth targeted validation. Not demonstrated clinical toxicity. |
| HORMAD1, DAZL | Neither / Cellular | Repeated immune-RNA observations make a simple rounding rescue less persuasive; examine donors, mapping and protein. |
| ACRBP, TEX101 | Unsafe / High RNA | Multiple immune observations exceed 30 in source-native units. Keep excluded under the current policy pending source-level adjudication. |
| FOXR2 | Neither / Antibody | Strong biological nomination but unresolved antibody conflict. Valid affirmative protein/recognition evidence can address low bulk RNA; it cannot erase uncertain normal staining. |
| CTAG2 | Neither / Variable | Keep the reviewed donor-dependent vital-tissue concern separate from a claim of proven myocardial toxicity. |
| XAGE5 | Neither / Antibody | No gene-specific rescue follows from XAGE1 family interest; neural staining and low reproductive RNA remain unresolved. |
| MAGEA3 | Unsafe / High RNA | Current colon-cell RNA exclusion is separate from historic receptor cross-reactivity with titin or MAGE-family proteins. |
| MAGEA12 | Unsafe / Safety | Retain the independent neurological-safety hold; RNA-threshold relaxation is insufficient. |
| SPAG9 | Unsafe / Protein | Published cancer antigenicity does not override broad normal protein and immune RNA. |
All threshold-boundary genes, including ones not named here, are searchable in the explorer. A near-boundary result is a reason to review measurement quality, not an automatic exemption.
What the rules still miss
- Published normal protein has not been comprehensively curated. FTMT is a concrete example of old tissue summaries missing later human protein data. The new decisions belong in the existing reviewed normal-protein mechanism, with assay, species, compartment, identity and contradictory evidence attached.
- Rare cells and biological states. Bulk averages dilute neurons, ductal cells, keratinocyte differentiation states and activated immune populations. Current immune/corneal/HPA cell sources help, but no negative baseline atlas proves silence during inflammation, repair or treatment.
- A high reproductive maximum can mask diffuse normal expression. Passing relative enrichment is not evidence of narrow distribution. Some retained genes have substantial normal cell RNA below the 30 cutoff. Bulk ceilings and cell-state review rules are distinct; a cell nTPM value must not silently be treated as a matched bulk measurement. Review the complete distribution and coverage, not only a single maximum or a thresholded count of tissues.
- Normal peptide presentation is missing from this gene-level audit. For TCR targeting, endogenous peptide/HLA evidence is closer to the intervention than total RNA or IHC. The HLA Ligand Atlas primary study provides a benign-tissue reference; lack of detection remains limited by tissue/HLA coverage and assay sensitivity. This audit has not ingested that dataset or certified its absence calls for these candidates.
- Therapy evidence is product-specific. PRAME's IMA203 study includes tumor and normal peptide-presentation work and clinical outcomes; it supports that defined peptide/receptor context, not every PRAME-derived sequence. Primary trial MAGE-A4-directed treatment similarly supports a defined clinical use, not automatic acceptance of every disputed antibody stain or all MAGE paralogs. Primary trial
- Cancer occurrence and actionable target abundance are separate. Of the 399 retained genes, the existing report shows at least one cancer RNA sample at ≥5 pTPM for 392; only 124 have Medium/High cancer IHC in the displayed matched groups. Another 107 have matched cancer IHC without such staining, and 168 lack exact-ID cancer IHC coverage. A single positive tumor is not broad indication coverage, clonality, peptide presentation or efficacy.
- Gene, proteoform and epitope decisions differ. Identical XAGE1A/B or CTAG1A/B proteins do not provide locus-specific expression proof. Different untranslated transcripts can encode the same protein. Exact k-mer exclusions cannot by themselves rule out TCR recognition of a different peptide.
Recommended next curation pass
Apply the five proposed normal-protein exclusions and three provisional holds through the existing reviewed-evidence workflow after inspecting the recorded limitations. Then prioritize the Vital antibody discrepancies, isoform conflicts and boundary exclusions. Preserve a candidate-level status until normal protein, presented peptide and product-specific safety have been adequately assessed.
Do not tune a universal threshold merely to preserve a familiar CTA name. Keep the literature conflict, measurement uncertainty and any proposed rescue visible alongside the unchanged released decision.