start / graph / ana-q180-the-af4-specific-definition-throws-away-aff1-aff4-and-ell

AnalysisstrongQ180-the-af4-specific-interactors

"Seen in MLL-AF4 and no other bait" excludes AFF1, AFF4 and ELL — the three most AFF1-specific proteins in the experiment — and keeps ACADM and GLUD1

also called AF4-specific definition critique · Skucha singleton class definition · n_fusions == 1

A published table lists 110 proteins that stuck to the MLL-AF4 fusion and to none of the six other fusions tested, and those 110 have been treated as the partners unique to our fusion. But the rule that built that list throws out AFF1 itself — the very protein our fusion is half made of — because a trace of it also turned up in one other pulldown, while it keeps two mitochondrial enzymes.

The same thing, for a scientist

In Skucha 2018 (PMID 29777171, Jurkat T-ALL, Strep-HA-tagged single-copy MLL-fusion transgenes) the operational definition of "specific to bait X" is `n_fusions == 1`: the protein appears in the top-300 list of one bait and no other. Applied to MLL-AF4 this yields 110 proteins. **It excludes AFF1 (MLL-AF4 p = 5.4e-43, 98 spectra), AFF4 (p = 3.83e-14, 48 spectra) and ELL (p = 5.52e-05, 19 spectra)** — the AFF1/SEC axis, i.e. everything a biologist would call AF4-specific — because each was also recovered under MLL-AF9, where AFF1 and AFF4 are genuine SEC components. Meanwhile the strongest member of the retained 110 is RGPD8 at p = 0.0147 on 3 spectra. The definition is a set operation over seven columns of one experiment; it is not a statement about AFF1 biology, and its sensitivity to a single extra column is absolute — one extra spectrum in the MLL-AF9 lane would have demoted any member of the 110, and one fewer under MLL-AF9 would have promoted AFF1 to the top of it.

Go deeper

The exclusion, verbatim from the sheet

Read back from analyses/skucha-mll-fusion-interactome/cache/41467_2018_4329_MOESM4_ESM.xlsx with the column map asserted against the sheet's own header labels 1:

protein	n_fusions	MLL-AF4 p	MLL-AF4 spectra	also recovered in	counted as AF4-specific
AFF1	2	5.4e-43	98	MLL-AF9 (p 3.5e-05, 35 spectra)	NO
AFF4	2	3.83e-14	48	MLL-AF9 (p 2.1e-10, 66 spectra)	NO
ELL	2	5.52e-05	19	MLL-AF9 (p 0.027, 13 spectra)	NO
MLLT1 (ENL)	4	1.31e-10	49	AF9, CBP, ENL	NO
RGPD8	1	0.0147	3	—	YES
ACADM	1	0.0208	8	—	YES
GLUD1	1	0.0399	11	—	YES

AFF1 is the C-terminal moiety of the bait itself. Its recovery at p = 5.4e-43 on 98 spectra is the experiment working; its recovery under MLL-AF9 as well is the SEC working. The definition reads both as evidence against AF4-specificity.

Four of the seven bait moieties ARE in their own specific list, which is the calibration

The n_fusions == 1 class is not noise — it contains each fusion's own 3' moiety wherever that moiety is not shared, and those are the best available positive controls for the axis 2:

bait	its 3' moiety	n_fusions	p in its own pulldown	spectra	in the specific list
MLL-CBP	CREBBP	1	1.77e-55	121	yes
MLL-AF1p	EPS15	1	1.58e-45	95	yes
MLL-EEN	SH3GL1	1	1.08e-23	257	yes
MLL-GAS7	GAS7	1	3.13e-17	183	yes
MLL-AF9	DOT1L (not the moiety, the canonical partner)	1	0.00172	24	yes

So when this axis detects real fusion-specific biology it does so at p between 1e-17 and 0.0017 on 24 to 257 spectra. The best of all 110 MLL-AF4 singletons is p = 0.0147 on 3 spectra — an order of magnitude weaker than DOT1L and forty-odd orders weaker than the moiety self-hits. The scale on which the class can be believed is visible inside the class itself, and the six p<0.05 MLL-AF4 members are not on it.

What would make a member of the 110 credible

Not a lower p-value inside this table. The three things that would move a member of this list are (i) recovery in a second, independent MLL-AF4 pulldown in a B-lineage host, (ii) peptide-level evidence from PXD009338 showing the identification does not rest on shared or borderline peptides, and (iii) a directed assay — co-IP, BioID or a structural contact. None exists for any of the 110. See exp-q180-reanalyse-pxd009338-for-peptide-level-evidence-under-the-rank-cut for the cheapest of the three.

Facts and figures

script	analyses/q180-is-af4-specificity-a-depth-artefact/positive_controls.py
status	run
result summary	The AF4-specific set (n_fusions == 1, n = 110) excludes AFF1, AFF4 and ELL, the three AFF1/SEC-axis proteins, each on the grounds of a second recovery under MLL-AF9. Four of seven bait 3' moieties and DOT1L are inside their own specific sets at p 1e-17 to 0.0017 on 24-257 spectra; the best of MLL-AF4's 110 is p 0.0147 on 3 spectra.
shown for fusion	YES for MLL-AF4 as one of seven baits in a single experiment — but a Strep-HA-tagged single-copy transgene in a Jurkat T-ALL host, not the endogenous KMT2A::AFF1 in B-lineage disease. See ana-q180-the-only-mll-fusion-interactome-is-a-tagged-transgene-in-a-t-all-host.
epigenetic readout	none - this is a proteomic set operation; no histone mark, occupancy or chromatin measurement is involved.
relationship to existing entries	Extends ana-the-index-fusion-interactome-is-not-distinctive-and-two-earlier-answers-were-artifacts, which already established that the table is rank-selected per bait and that MLL-AF4's 110/291 (37.8%) uniqueness is a depth artefact. That entry did NOT notice that the same definition excludes AFF1, AFF4 and ELL, and did not calibrate the singleton class on its own positive controls. Nothing here contradicts it.

measurements

inputs

[
  "analyses/skucha-mll-fusion-interactome/cache/41467_2018_4329_MOESM4_ESM.xlsx (Skucha 2018 Supplementary Table 5)"
]

measurements — the numbers behind the chart

[
  {
    "quantity": "p_value",
    "subject": "AFF1|MLL-AF4",
    "value": 5.4e-43,
    "units": null,
    "n": 2,
    "universe": "Skucha 2018 Suppl Table 5, GLM over spectral counts, top-300 per bait",
    "null": null,
    "note": "98 spectral counts summed over two replicates; excluded from the AF4-specific set because n_fusions = 2"
  },
  {
    "quantity": "p_value",
    "subject": "RGPD8|MLL-AF4",
    "value": 0.0147,
    "units": null,
    "n": 2,
    "universe": "Skucha 2018 Suppl Table 5, same GLM",
    "null": null,
    "note": "3 spectral counts; the STRONGEST of the 110 proteins the same table marks AF4-specific"
  },
  {
    "quantity": "p_value",
    "subject": "DOT1L|MLL-AF9",
    "value": 0.00172,
    "units": null,
    "n": 2,
    "universe": "Skucha 2018 Suppl Table 5, same GLM",
    "null": null,
    "note": "24 spectra; the only non-trivial true positive inside the singleton class, and the calibrator this entry uses"
  },
  {
    "quantity": "count",
    "subject": "bait 3' moieties inside their own singleton set",
    "value": 4,
    "units": null,
    "n": 7,
    "universe": "the seven MLL-fusion baits of Skucha 2018",
    "null": null,
    "note": "CREBBP, EPS15, SH3GL1, GAS7 yes; AFF1, MLLT3, MLLT1 no"
  }
]

How this could bite us back

If a future beat proposes a drug or a degrader against a member of the 110 on the strength of 'AF4-specific interactor', the supporting evidence is a set operation over one experiment with a best p of 0.0147 on 3 spectra. That is not a target-validation package and it should not be presented as one.

Still open

What does the MLL-AF4 column look like if 'AF4-specific' is defined by EFFECT rather than by set membership — for instance, preys whose MLL-AF4 spectral count exceeds the maximum over the other six baits by more than 3-fold? AFF1 would be first under that definition. Nobody has computed it, and it is one pass over the same cached workbook.
The paper's body states that 406 of 960 network proteins (42.3%) co-purified with more than one MLL-fusion. The n_fusions column of Supplementary Table 5 gives 392 (40.8%). The >=5 figure (128) and the all-seven figure (42) both match the table exactly, so the discrepancy is confined to the '>1' count. Which of the two is the figure the network in Supplementary Fig. 2b was drawn from?
Is there any MLL-fusion AP-MS dataset other than PXD009338? If not, every 'fusion-specific partner' claim in this field rests on one laboratory, one cell line and one instrument.

Every source behind this entry (2)

E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
In Skucha 2018's Supplementary Table 5, AFF1 (MLL-AF4 p = 5.4e-43, 98 spectral counts summed over two replicates), AFF4 (p = 3.83e-14, 48) and ELL (p = 5.52e-05, 19) all carry n_fusions = 2 because each was also recovered under MLL-AF9, and are therefore excluded from the 110 proteins the table marks as recovered under MLL-AF4 alone; the strongest member of those 110 is RGPD8 at p = 0.0147 on 3 spectral counts.
MLL-fusion-driven leukemia requires SETD2 to safeguard genomic integrity. · Skucha A, Ebner J, Schmöllerl J, Roth M, et al. (20 authors; Grebien F senior author) · Nature Communications 2018
PMID 29777171 DOI
What this evidence cannot show
Read from the cached Supplementary Table 5 workbook (MOESM4) in this repository, not from the paper's prose; the numbers are in the supplement and do not appear in the abstract or body. The column map was asserted against the sheet's own header row before any value was read (`analyses/q180-is-af4-specificity-a-depth-artefact/positive_controls.py`). Spectral counts are the sum of the two replicate values the sheet prints as 'a|b'. The p-values are from the paper's own generalised linear model, not recomputed here.
Evidence that argues against this (1)
- The exclusion is not obviously wrong as a definition. AFF1 and AFF4 really were recovered under MLL-AF9 as well, and a set-difference over seven baits is a reasonable way to ask which preys are private to one bait. The defect is in how the resulting list is USED: it is a within-experiment contrast, and calling it 'the AF4-specific interactome' imports a biological claim the set operation cannot carry. A reader who wants AFF1-axis partners should query the MLL-AF4 column by p-value, not by n_fusions. Nature Communications 2018
E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
The n_fusions = 1 class contains genuine positives: CREBBP under MLL-CBP (p 1.77e-55, 121 spectra), EPS15 under MLL-AF1p (1.58e-45, 95), SH3GL1 under MLL-EEN (1.08e-23, 257), GAS7 under MLL-GAS7 (3.13e-17, 183) — each bait's own 3' moiety — and DOT1L under MLL-AF9 (0.00172, 24), the canonical partner-specific interactor; every one of these is stronger than every one of MLL-AF4's 110 singletons, whose best p is 0.0147 on 3 spectra.
Positive-control calibration of the Skucha 2018 singleton class, run 2026-09-24 by Q180. Script analyses/q180-is-af4-specificity-a-depth-artefact/positive_controls.py, output positive_controls.json; column map asserted against the sheet's header row. · This repository (script and JSON committed; deterministic, no network) 2026
What this evidence cannot show
MLLT3 (AF9) and MLLT1 (ENL) are NOT in their own specific lists (n_fusions 2 and 4), so the moiety self-identification control is available for four of seven baits, not seven of seven. MLLT1 under MLL-ENL carries p = 9.92e-23 on only 3 spectra, which shows the p-value in this table is not a monotone function of spectral count — the GLM's baseline term can drive an extreme p from few spectra when the control baseline is zero. That cuts both ways and is why this entry quotes p AND spectra everywhere.
Evidence that argues against this (1)
- A bait's own 3' moiety is a trivially-true positive and may calibrate the axis too generously: it is present at stoichiometry 1:1 with the bait by construction, so 95-257 spectra tells you nothing about the scale on which a genuine sub-stoichiometric partner would appear. DOT1L at 24 spectra is the more honest calibrator, and it is the only non-trivial one available. If the true scale for a real partner-specific interactor were nearer 5 spectra than 24, the six MLL-AF4 targets would not be excluded by this argument. Nothing in this table settles that. This repository 2026

Connected to

← supportsCORRECTED: the AF4-specific interactome gap is ZERO, not one - RPA2 was declined on purpose in cycle 159 and my no-match read was the tool warning I ignored
composes with the depth-artefact result to show the singleton class cannot be read as 110 leads
supports →Three quarters of KMT2A::AFF1's top partners are shared with other MLL fusions — and the two analyses that said otherwise were both artifacts of how the source table was built
extends the depth-artefact finding with the definitional exclusion of AFF1/AFF4/ELL and a positive-control calibration the earlier entry did not run
related to →AFF1 / AF4 — the 3' half of the fusion, and a target with almost no structural footing
AFF1 is recovered under MLL-AF4 at p 5.4e-43 on 98 spectra and is excluded from the AF4-specific set by the n_fusions == 1 rule
related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
MLL-AF4 in Skucha 2018 is the index fusion; this adjudicates what its 'specific' partner list can and cannot mean
related to →DOT1L — the target where excellent chemistry met an indirect interaction
DOT1L under MLL-AF9 (p 0.00172, 24 spectra) is the only non-trivial true positive inside the singleton class and is used here as its calibrator
← related toAcross all seven MLL-fusion baits, how deep a bait's fixed top-300 list runs into non-significance predicts how many "specific" partners it appears to have — Spearman +0.821, exact p = 0.034, and MLL-AF4 is the extreme on both axes
the depth confound explains why the AF4 singleton class is large; the definition critique explains why it contains the wrong proteins
← depends onRGPD8 is the top-ranked MLL-AF4-specific prey on 3 spectra, and 96.5% of its tryptic peptides are shared with seven near-identical paralogues — only 4 peptides in the whole protein could ever have discriminated it — declined
RGPD8 is the strongest member of the retained 110 at p 0.0147 on 3 spectra, the comparison that entry uses to calibrate the class
← depends onCEP112 is a 955-residue centrosomal coiled-coil protein on 7 spectra with zero leukaemia literature — its only nuclear annotation is a single 2024 citation, and a long coiled-coil is the classic non-specific AP-MS passenger — declined
sits at rank 95 of 291, well inside the region the calibration shows is non-significant
← depends onKMT2A::AFF1 has no measured fusion-specific interactome: the one dataset that appears to supply one is a rank-selected tail in the wrong lineage, and the question is open rather than answered badly
the definitional exclusion is the fourth measurement the hypothesis rests on
supports →ACADM is a mitochondrial matrix beta-oxidation enzyme with an experimentally coded matrix-only localisation, recovered on 8 spectra from a nuclear-extract pulldown whose specific tail is 14.5% mitochondrial — declined
ACADM is one of the two mitochondrial enzymes the definition retains while excluding AFF1
supports →GLUD1 is a hexameric mitochondrial matrix dehydrogenase recovered on 11 spectra with non-zero control signal from a nuclear pulldown — and this corpus has already declined one KMT2A partner whose sole source was a cryo-EM structure of GLUD1 — declined
GLUD1 is the other mitochondrial enzyme the definition retains while excluding AFF1, AFF4 and ELL