Oncofusion

start / graph / ana-the-expression-control-splits-the-verdict

Analysisunrated confidence 0 of 5: unrated

The paralogue split survives the expression control and the ALAS2 control does not: ING4 is transcribed MORE than ING5 and is not required at all

also called expression control paralogue split · ALAS2 artefact

A gene that a cell never switches on will look unnecessary in any experiment that switches genes off, so we checked whether the genes that looked unnecessary were switched on. Most were. One was not, and that one was being used as evidence.

The same thing, for a scientist

Two cycle-60 findings shared one unexamined assumption - that a low-ranking gene is present and not needed, rather than simply absent. Checked against Human Protein Atlas cell-line RNA across the seven panel lines HPA holds. **The paralogue split is real**: every low-ranking paralogue is expressed, and **ING4 is transcribed at a median 35.2 nTPM against ING5's 27.9** while ranking 10,776 to ING5's 148. **The ALAS2 control is an artefact**: ALAS2 sits at **0.10 nTPM** in all seven lines, so its rank of 9,992 records which isoform these cells transcribe and nothing about specificity. That half of `ana-depmap-haem-lineage-control` is withdrawn.

Go deeper
What was at stake, and why it needed an outside resource

A gene that is not transcribed scores as not essential in a CRISPR screen whatever its biology. So 'paralogue-specific selectivity' and 'only one paralogue is expressed' are the same data until expression is checked, and both cycle-60 entries named this as their first open question.

The check uses Human Protein Atlas cell-line RNA rather than DepMap's own expression, so the measurement is not answering for itself. HPA holds 7 of the 9 patients in this project's eleven-line panel: SEM, RS4;11, MV4-11, THP-1, NOMO-1, MOLM-13 and OCI-AML-2. It does not hold SHI-1 or HB11;19; SEMK2 and MOLM-14 are the same patients as SEM and MOLM-13.

The paralogue split is real, and ING4 is the clean case
familygeneselectivity rankmedian nTPMexpressed?
ING scaffoldING514827.9yes
ING410,77635.2yes - MORE than ING5
JADEJADE23815.8yes
JADE311,6354.5yes
JADE179624.6yes
acetyltransferaseKAT79223.1yes
KAT6A10620.8yes
KAT6B8,3394.7yes
BRPF/BRDBRPF125822.7yes
BRPF34,1217.9yes
BRD13,84521.5yes

Every low-ranking paralogue is expressed in every one of the seven lines. The artefactual reading is excluded and the finding stands.

ING4 is the case that carries the argument. It is transcribed at a higher median level than its selective sibling ING5 - 35.2 against 27.9 nTPM - and ranks 10,776 of 17,728, with a gene effect that is fractionally positive. Abundant, present in every line, and not required at all. No expression argument can touch that.

The weaker cases are KAT6B and JADE3. Both are expressed, but at roughly a fifth of their selective siblings (4.7 against 20.8; 4.5 against 15.8). 'Expressed' is not the same as 'expressed enough to substitute', so for those two families a dosage explanation survives this check even though the absence explanation does not. BRD1 at 21.5 nTPM and rank 3,845 is a second clean case, and ING4 the cleanest.

The ALAS2 control is withdrawn

ana-depmap-haem-lineage-control presented the ALAS1/ALAS2 separation as an internal specificity control: two isoforms of the same enzymatic step, ALAS1 at rank 47 and ALAS2 at 9,992, which would be hard to explain if the pathway signal were artefactual.

ALAS2 sits at 0.10 nTPM in all seven lines. It is not expressed.

So the separation records which isoform these cells transcribe - the erythroid-specific one is silent in myeloid and lymphoid lines, as it should be - and carries no information about whether the haem signal is specific. That entry's own counter-evidence note predicted exactly this and it turned out to be right, which is the argument for writing counter-evidence that can be checked rather than counter-evidence that sounds balanced.

The rest of the haem result is untouched. All eight canonical enzymes are expressed at 7.9 to 67.0 nTPM, so the pathway-level finding - median rank 115 of 17,728, 0 of 20,000 random draws - is not an expression artefact. What it loses is a control it never had.

Still open
  1. Is the ING5-versus-ING4 difference functional at the protein level? ING4 and ING5 are the swapped subunit between HBO1-4.x and HBO1-5.x complexes, and ING4 being abundant and dispensable while ING5 is required is a specific, testable claim about which complex matters.
  2. Does KAT6B or JADE3 rescue when overexpressed to KAT6A or JADE2 levels? That is the experiment that separates the dosage reading from the function reading for the two families where both survive.
  3. HPA's lines are not the DepMap isolates. Does DepMap's own expression data agree? It would be a same-source check and weaker, but disagreement between the two would matter more than agreement.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis

    Every low-ranking paralogue in the HBO1/MOZ acetyltransferase modules is expressed in all seven KMT2A-rearranged panel lines held by the Human Protein Atlas. ING4 is transcribed at a median 35.2 nTPM against ING5's 27.9 while ranking 10,776 against ING5's 148; KAT6B 4.7, JADE3 4.5, BRPF3 7.9 and BRD1 21.5 nTPM are all above the detection threshold.

    What this evidence cannot show

    Run 2026-09-20 over 24,315,372 rows covering 1,206 cell lines. **HPA holds 7 of the 9 patients in the panel, not all of them** - SHI-1 and HB11;19 are absent - and HPA's lines are its own cultures, not the DepMap isolates the gene effects come from, so this is a different laboratory's stock of a line with the same name. **RNA is not protein**, and KAT6B and JADE3 are expressed at roughly a fifth of their selective siblings, which leaves a dosage explanation open for those two families even though absence is excluded.

    Evidence that argues against this (2)
    • **The dosage objection survives for two of the four families.** At 4.5 and 4.7 nTPM, JADE3 and KAT6B may be present but too scarce to substitute for JADE2 and KAT6A, which would make the split a quantitative rather than a qualitative fact about paralogue function. Only ING4 - more abundant than its selective sibling and completely dispensable - and BRD1 at 21.5 nTPM are immune to that reading. Oncofusion knowledge graph (internal analysis) 2026
    • A CRISPR gene effect near zero can also mean the guides failed, the protein is buffered by something other than its paralogue, or the knockout is tolerated over the screen's timescale. Expression excludes one explanation of a null, not all of them. Human Protein Atlas 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis

    ALAS2, the erythroid-specific isoform used as an internal specificity control for the haem pathway finding, is not expressed in any of the seven panel lines - median 0.10 nTPM. Its selectivity rank of 9,992 against ALAS1's 47 therefore records which isoform these cells transcribe rather than any specificity of the haem signal.

    What this evidence cannot show

    0.10 nTPM is HPA's floor for these lines rather than a measured trace, so the honest statement is 'at or below detection' rather than 'exactly 0.10'. This withdraws a control, not the finding it was attached to: the other eight haem enzymes are expressed at 7.9 to 67.0 nTPM and the pathway-level result stands.

    Evidence that argues against this (1)
    • The withdrawal was predicted in the entry that made the claim, whose own counter-evidence note said ALAS2 scoring low 'is exactly what ALAS2 is not expressed in these lines predicts'. That the prediction was recorded and then confirmed is the process working; that the control was published before the check was run is not. Oncofusion knowledge graph (internal analysis) 2026
Connected to