Oncofusion

start / graph / ana-the-acetyltransferase-signal-is-paralogue-specific-and-does-not-survive-pan-essentiality

Analysisunrated confidence 0 of 5: unrated

The HBO1/MOZ modules are not selective as modules: one paralogue per family carries the signal, and no member clears both a dependency and a pan-essentiality bar

also called acetyltransferase paralogue split · HBO1 module per-gene test

A sweep of 719 protein machines put a family of histone-modifying machines at the top for this leukaemia. Tested gene by gene, the signal turns out to come from one specific member of each family while its near-identical siblings do nothing - and none of them is both important and specific enough to be a target.

The same thing, for a scientist

`ana-the-complex-level-sweep-kills-the-coherence-mechanism` found 8 of the 12 most KMT2A-selective complexes were HBO1/MOZ/MORF modules and flagged the n=4 median as suspect. Tested per gene: **selectivity splits by paralogue with rank spreads of 3,863 to 11,597** - JADE2 rank 38 against JADE3 11,635; ING5 148 against ING4 10,776; KAT6A 106 against KAT6B 8,339. **'The complex is selective' is the wrong description.** And against a pan-essentiality control, **no module gene clears both bars**: the three with the largest KMT2A-rearranged effects (KAT7, MEAF6, BRPF1) are broadly essential, and the three that are not broadly essential (KAT6A, ING5, JADE2) never reach a dependency-strength effect. Only **34 of 17,728 genes** clear both, and PSIP1/LEDGF is 3rd among them.

Go deeper
The paralogue split, which the complex median cannot see
familymembers, by selectivity rankspread
JADE scaffoldJADE2 38, JADE1 796, JADE3 11,63511,597
ING scaffoldING5 148, ING4 10,77610,628
acetyltransferaseKAT7 92, KAT6A 106, KAT6B 8,3398,247
BRPF/BRD readerBRPF1 258, BRD1 3,845, BRPF3 4,1213,863

Every family splits the same way: one member near the top of 17,728, its paralogues nowhere. That is not what a module-level dependency looks like.

And the complex-level statistic actively hid it. Complex Portal lists HBO1-4.2 and HBO1-5.2 as variants differing in exactly one subunit - ING4 against ING5. HBO1-4.2 contains ING4 at rank 10,776, and still ranked 3rd of 719, because its median over four members is taken from ING4 (10,776), JADE2 (38), KAT7 (92) and MEAF6 (198) - a median of 145, with the one uninformative member sitting outside it. The median of four hides the single discriminating subunit by construction, which is a sharper version of the small-complex caveat the sweep already carried.

This also sits in tension with hyp-paralogue-buffering-hides-dependencies, which argues that single-gene knockout understates dependencies because paralogues cover for each other. Here the paralogues plainly do not cover: JADE3 is not rescuing JADE2. Both can be true of different families, but this is a measured case against the general form.

The pan-essentiality control, which none of them survives

A gene that every cell line needs is not a selective dependency however large its KMT2A-rearranged effect. Applying two bars - a real dependency in rearranged lines (mean Chronos <= -0.50) and not broadly essential (mean across all 1,208 lines > -0.20):

genemean in KMT2A-rpan-essentialityverdict
KAT7-0.650-0.300strong, but broadly essential
MEAF6-0.653-0.356strong, but broadly essential
BRPF1-0.544-0.262strong, but broadly essential
KAT6A-0.429-0.095specific, but not a dependency
JADE2-0.389-0.006specific, but not a dependency
ING5-0.245+0.028specific, but not a dependency

The split is perfect and it goes the wrong way. Every member with a dependency-strength effect is broadly essential; every member that is specific has a weak effect. No module gene clears both bars.

For scale, 34 of 17,728 genes clear both. The top of that list is FLT3 (-0.84), SLC5A3 (-0.74), PSIP1/LEDGF (-0.56, pan-essentiality +0.03), MEF2C, MED13L and IRF8. MEN1 is not on it - its pan-essentiality is -0.363, so by this criterion menin is a strong but broadly-needed gene rather than a clean selective one, which is a real qualification on the target this project is named for. DOT1L is 25th of the 34.

PSIP1 at 3rd is the result worth carrying forward. LEDGF is essentially dispensable across DepMap (+0.028) and a genuine dependency in KMT2A-rearranged lines (-0.553). That is the profile the acetyltransferase module conspicuously lacks, and it is independent support for the LEDGF arm of the menin-interface work from a scan that was run without a prior.

Still open
  1. Does KAT6B or JADE3 rescue when overexpressed to KAT6A or JADE2 levels? Expression is settled - both are present - so dosage against function is now the only surviving alternative for those two families, and one overexpression experiment separates them.
  2. PSIP1/LEDGF is the cleanest selective dependency the scan found after FLT3 and SLC5A3, and this graph's standing LEDGF gap is that no published experiment removes LEDGF from an established human KMT2A::AFF1 leukaemia. The computational case is now strong enough that the gap is the bottleneck.
  3. What is SLC5A3 doing at rank 3? A myo-inositol transporter with a -0.74 selective difference and no mention anywhere in this graph.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis

    Within each of the four paralogue families making up the HBO1/MOZ/MORF acetyltransferase modules, KMT2A-selectivity is carried by one member while its paralogues rank in the middle of the genome: JADE2 rank 38 against JADE3 11,635; ING5 148 against ING4 10,776; KAT6A 106 against KAT6B 8,339; BRPF1 258 against BRPF3 4,121.

    What this evidence cannot show

    Computed 2026-09-20 from the cached genome-wide scan; no new data. Every rank here comes from **11 KMT2A-rearranged lines against 43 other leukaemia lines**, and those 11 lines are **nine patients, seven myeloid, none infant-derived**. **Expression has now been checked and the artefactual reading is excluded.** Against Human Protein Atlas cell-line RNA across the seven panel lines HPA holds, every low-ranking paralogue is expressed: ING4 at a median 35.2 nTPM - *higher* than the selective ING5 at 27.9 - and KAT6B 4.7, JADE3 4.5, BRPF3 7.9, BRD1 21.5. What remains open is dosage rather than absence: KAT6B and JADE3 sit at roughly a fifth of their selective siblings, so for those two families a quantitative explanation survives.

    Evidence that argues against this (1)
    • The expression check excludes absence, not dosage. **ING4 is the only member that is immune to a dosage reading** - abundant, in every line, and fractionally *positive* in gene effect - with BRD1 at 21.5 nTPM a second clean case. For KAT6B and JADE3 at about a fifth of their siblings' transcript level, 'present but too scarce to substitute' remains a live explanation and would make the split quantitative rather than functional. Oncofusion knowledge graph (internal analysis) 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis

    No gene in the HBO1/MOZ/MORF modules is both a dependency in KMT2A-rearranged lines (mean Chronos <= -0.50) and not broadly essential (pan-essentiality > -0.20). KAT7, MEAF6 and BRPF1 have the strongest effects and are broadly essential; KAT6A, JADE2 and ING5 are not broadly essential and have sub-threshold effects.

    What this evidence cannot show

    The two thresholds are conventions, not facts: -0.50 is the usual Chronos dependency line and -0.20 is chosen here as a pan-essentiality bar. JADE2 at -0.389 and KAT6A at -0.429 fail the dependency bar by a margin smaller than the spread across eleven lines, so a different threshold changes the verdict for them. This is a screening-grade filter, not a decision.

    Evidence that argues against this (1)
    • Being broadly essential does not disqualify a target - it raises the therapeutic index question rather than settling it, and several approved drugs hit broadly essential proteins. The stronger reading is comparative: whatever KAT7 is, it is not the kind of selective dependency this scan was built to find, and the complex-level sweep presented it as though it were. Oncofusion knowledge graph (internal analysis) 2026
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis

    Only 34 of 17,728 scanned genes are both a dependency in KMT2A-rearranged lines and not broadly essential. PSIP1/LEDGF is 3rd on that list with a mean effect of -0.553 in rearranged lines and a pan-essentiality of +0.028. MEN1 is absent from the list because its pan-essentiality is -0.363.

    What this evidence cannot show

    MEN1's absence is an artefact of a threshold, not a finding about menin: it has the largest selective difference in the entire scan at -0.866 and ranks 1 of 17,728. What the pan-essentiality figure says is that menin is also needed by non-rearranged lines, which is already documented in this graph's endocrine and normal-tissue work. Reading its absence here as evidence against menin would be wrong.

    Evidence that argues against this (1)
    • A filter that excludes the target with the single largest selective effect in the scan is demonstrating its own crudeness. Pan-essentiality and selectivity are two axes and collapsing them into a pass/fail loses the ordering that matters; the 34-gene list should be read as a shortlist with a known blind spot, not as a ranking of targets. Oncofusion knowledge graph (internal analysis) 2026
Connected to