Excluding spleen, lymph node and thymus moves 9 of 29 normal-tissue safety ratios and not one of the worst cases - the artefact is real, concentrated and confined to the middle of the table, and it takes the ACTB reference down with it
also called lymphoid-excluded re-ranking · CD19 collapses 12.0 to 0.63 · PSIP1 was 22% thymus · the housekeeping baseline itself was inflated
A gene can look dangerously active in an organ simply because that organ is full of immune cells. Redoing the safety ranking with the three most lymphocyte-rich organs taken out changes nine genes out of twenty-nine, but leaves every one of the worst offenders exactly where it was.
The same thing, for a scientist
Cycle 136 found the leukocyte artefact runs opposite to the direction I had pre-registered: lymphocytes inside an organ **inflate** a haematopoietic antigen's highest-tissue/bone-marrow ratio. `concept-the-leukocyte-artefact-inflates-a-haematopoietic-antigens-tissue-ratio-it-does-not-deflate-it` named the fix and the cycle-136 log carried it as an open item. **This is that re-run.** The maximum is taken twice - once over every tissue but bone marrow, as published, and once additionally excluding **spleen, lymph node and thymus**. Bone marrow stays the denominator because it is the compartment being treated. 🔵 **The re-run reproduces the published numbers first: 13 of 13 within tolerance** (MEIS1 17.21 against 17.2, MEF2C 11.57 against 11.6, MLLT1 6.56 against 6.6, MEN1 3.26 against 3.3, ACTB 1.21 against 1.2). So this is a faithful re-execution and not a new metric [E1]. **The pre-registered R2 branch fires, and R3 explicitly does not.** Nine of 29 genes have a lymphoid top tissue, and **all nine of the genes that change are exactly those nine** - the change is the artefact, not panel composition. | gene | top tissue (all) | ratio | top tissue (parenchyma) | ratio | drop | |---|---|---:|---|---:|---:| | **CD19** *(control)* | spleen | 12.00 | testis | **0.63** | **-95%** | | **KDM6A** | thymus | 0.79 | ovary | 0.56 | -29% | | **KDM4C** | thymus | 0.77 | parathyroid | 0.59 | -23% | | **ACTB** *(reference)* | lymph node | 1.21 | lung | **0.94** | -22% | | **PSIP1 / LEDGF** | thymus | 2.07 | cerebral cortex | 1.62 | -22% | | CCNT1 | thymus | 0.44 | thyroid | 0.40 | -9% | | BOD1L1 | spleen | 0.98 | ovary | 0.92 | -6% | | RBBP5 | thymus | 1.38 | kidney | 1.30 | -6% | | KMT2A | thymus | 0.97 | ovary | 0.92 | -5% | 🔴 **CD19 collapses from 12.0 to 0.63, a 95% fall.** The control behaves exactly as the concept predicts, which is the strongest available evidence that the correction is doing what it claims. 🔵 **But the headline conclusions are robust, and that is the practically important half.** **Not one of the worst cases moves at all**: LAMP5 19.57, MEIS1 17.21, MEF2C 11.57, MLLT1 6.56, **PROM1 5.27** and MEN1 3.26 all have parenchymal top tissues - cerebral cortex, pancreas, kidney, liver - and are identical under both rules. **Every safety statement this project has published about its worst offenders survives unchanged.**
Two rows that matter more than their size
PSIP1 / LEDGF, 2.07 -> 1.62. This is not a bystander: LEDGF is part of the menin-LEDGF-KMT2A interface this whole programme is built around. Its apparent off-compartment exposure was 22% thymic lymphocytes, and the corrected figure puts it closer to the marrow-selective band. A small absolute change on a gene that is central to the mechanism.
🔴 ACTB, 1.21 -> 0.94, and this one undermines the scale itself. ACTB is the reference point the published entry used to say what "normal" looks like on this metric - and its top tissue was lymph node. So the yardstick was inflated by the same artefact as the genes being measured. Corrected, ACTB is below 1.0, meaning a ubiquitously expressed housekeeping gene is more abundant in bone marrow than in any parenchymal organ on this panel. That is the right answer for marrow, which is highly transcriptionally active, and it means the band for "unremarkable" sits lower than the published table implies.
Why this is a blunt instrument, and what would be better
Excluding whole organs is crude in both directions. Parenchymal organs contain resident immune cells too - lung and intestine especially - so the parenchyma-only maximum is still contaminated, just less. And spleen has genuine non-lymphoid function, so for a gene with real splenic biology the exclusion discards a true signal.
The concept entry's Option 2 - deconvolution against an immune-cell signature - remains the better fix and still needs a reference this project does not hold. This run implements Option 1 because it is two lines and testable today.
A direction was declared in advance and asserted in code: excluding tissues can only lower the numerator, so every ratio must move down or stay equal, and a rise would be a bug rather than a finding. None rose. That assertion exists because cycle 136's escape condition was unfireable for want of exactly this kind of stated direction.
| script | analyses/does-excluding-lymphoid-organs-change-the-safety-ranking/run.py |
|---|---|
| status | run |
| run date | 2026-09-24 |
| n genes | 29 |
| denominator | bone marrow, unchanged |
| regression vs published | 13 of 13 within tolerance |
| n lymphoid top | 9 |
| n changed | 9 |
| n dropping 20pct or more | 5 |
| preregistered branch fired | R2 - the ranking moves; R3 explicitly does not, the change is concentrated on the lymphoid-top set |
| declared direction held | no ratio rose; excluding tissues can only lower the numerator, asserted in code |
lymphoid excluded
[ "spleen", "lymph node", "thymus" ]
worst cases unchanged — the numbers behind the chart
{
"LAMP5": 19.57,
"MEIS1": 17.21,
"MEF2C": 11.57,
"MLLT1": 6.56,
"PROM1": 5.27,
"MEN1": 3.26
}notable moves
{
"CD19": [
12,
0.63
],
"PSIP1": [
2.07,
1.62
],
"ACTB": [
1.21,
0.94
]
}measurements
[
{
"quantity": "genes whose highest-tissue/bone-marrow ratio changes when lymphoid organs are excluded",
"value": 9,
"unit": "count",
"subject": "29 genes: the published 22 candidates, their 2 controls, and cycle 136's 5 surface antigens",
"universe": "a 16-tissue HPA consensus panel; 9 of 29 have a lymphoid top tissue and all 9 are the genes that change"
},
{
"quantity": "CD19 ratio, all-tissue versus parenchyma-only",
"value": {
"all": 12,
"parenchyma": 0.63
},
"unit": "ratio of nTPM",
"subject": "CD19, the positive control for the leukocyte artefact",
"universe": "the same panel; a 95% fall, which is the artefact behaving as the concept predicts"
}
]- Find an immune-cell deconvolution reference for HPA consensus tissues. Excluding whole organs is Option 1 and crude; deconvolution is Option 2 and still unavailable to this project.
- Is high thymic PSIP1 and KMT2A expression contamination or real developmental biology? For a paediatric indication the thymus is not obviously a tissue to discard, and treating it as an artefact may hide a genuine exposure.
- Re-express the whole table as the raw pair (bone marrow nTPM, worst parenchymal nTPM) rather than a ratio. CSPG4 and ALB return nothing under either rule, and the ACTB result shows the ratio's reference point is itself unstable.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.
Re-running the Human Protein Atlas normal-tissue filter over 29 genes reproduces 13 of 13 published ratios within tolerance, and excluding spleen, lymph node and thymus from the numerator changes exactly the 9 genes whose top tissue was lymphoid: CD19 12.00 to 0.63, KDM6A 0.79 to 0.56, KDM4C 0.77 to 0.59, ACTB 1.21 to 0.94, PSIP1 2.07 to 1.62, CCNT1 0.44 to 0.40, BOD1L1 0.98 to 0.92, RBBP5 1.38 to 1.30 and KMT2A 0.97 to 0.92, while LAMP5, MEIS1, MEF2C, MLLT1, PROM1 and MEN1 are unchanged.
What this evidence cannot show
Run 2026-09-24 (cycle 137) on the same 16-tissue panel and identifier guard as the two earlier runs, with exactly one exact-symbol row required per gene. CSPG4 and ALB have bone marrow 0.0 so both ratios are undefined for them under either rule and they are excluded from the comparison. Excluding whole organs is blunt: parenchymal organs contain resident immune cells, so the parenchyma-only maximum is still contaminated, and spleen has genuine non-lymphoid function that the exclusion discards. HPA consensus values are adult tissue.
Evidence that argues against this (2)
- That the changes fall entirely on the lymphoid-top set is close to tautological: a gene whose maximum is not in spleen, lymph node or thymus cannot change when those are removed. The informative content is therefore not the concentration but the SIZE of the drops and which genes are affected - and on that reading the result is modest, since only 5 of 29 lose 20% or more and four of those sit below a ratio of 2.1 where the absolute stakes are small.
- Treating the parenchyma-only figure as the corrected truth assumes the lymphoid signal is always contamination. For a target intended to act on lymphoid malignancy that assumption can be wrong in the other direction: high thymic expression of PSIP1 or KMT2A may be a real statement about a tissue where these genes matter developmentally, and discarding it hides a genuine paediatric exposure rather than removing an artefact.
runs the exclusion that concept proposed; the CD19 control falls 95%, confirming it
reproduces 13 of 13 of its ratios and supplies parenchyma-only figures for the nine lymphoid-topped genes
PROM1 5.27 is unchanged under the exclusion because kidney is parenchyma, so that reading survives
LAMP5 19.57 is unchanged; cerebral cortex is not a lymphoid organ
PSIP1/LEDGF drops 2.07 to 1.62; 22% of its apparent off-compartment exposure was thymic lymphocytes
- related to →KMT2A / MLL1 — the gene that breaks
KMT2A 0.97 to 0.92 under the exclusion; its top tissue was thymus