No orthogonal validation exists for any of the 110 MLL-AF4-specific partners, as far as four declared bounds can see — and the one partner the paper itself followed up, SETD2, is on the opposite list, shared by all seven fusions
also called no orthogonal validation of the AF4-specific tail · the 110 have never been followed up
The published list of 110 proteins said to bind only the MLL-AF4 fusion has never been checked by anyone, with any other method, in the eight years since it appeared. The one protein the authors did follow up, SETD2, was on the opposite list — the one shared by all seven fusions.
The same thing, for a scientist
Skucha 2018 followed its own interactome with a 40-gene shRNA screen and then with substantial functional work on **SETD2** — which is a **pan-fusion** partner, recovered under all seven baits (MLL-AF4 p = 0.0372, 3|4 spectra) [E1]. **No member of the 110 MLL-AF4-only set has been carried into any orthogonal assay by the authors or by anyone else.** Bounding this: IntAct has never curated PMID 29777171, so no curated interaction record exists for any of them [E2]; PubMed queries pairing each of this beat's six p<0.05 candidates with KMT2A, MLL or leukaemia return 0-2 records, none on topic [E3]; and the corpus-content join finds 89 of the 110 never mentioned anywhere in this graph, with 6 of the 21 that are mentioned appearing only in the entry that reported the gap [E4]. **`absence_class` is `measured-not-analysed`, not `never-measured`** — the primary data is public in PXD009338, so this is a follow-up that nobody has done rather than an experiment nobody can do.
What the paper validated, and why that is the opposite list
The paper's own filtering narrative, from its Fig. 2a legend: AP-MS identified 960 candidates (top 300 per bait); 128 proteins interacted with ≥5 of seven; 40 candidates were screened in MLL-rearranged versus MLL-wild-type lines 1. The screen was seeded from the SHARED end of the distribution, not the specific end. SETD2 emerged, and the rest of the paper is about SETD2 — co-IP, shRNA, survival, H3K36me3, genome integrity.
So the follow-up effort in the one paper that produced this dataset went entirely to the pan-fusion core. That is a defensible scientific choice — the authors were looking for a unifying mechanism across MLL fusions and said so in their first sentence — but it means the AF4-specific list was published and abandoned in the same paper, and this corpus's coverage pattern (42 of 42 shared, 1 of 110 specific) mirrors the source paper's own priorities rather than diverging from them.
What would close this absence, and what it would cost
| route | closes | cost |
|---|---|---|
| peptide-level parse of the 12 deposited SEARCH files | whether the identifications are sound | low, no wet work — exp-q180-reanalyse-pxd009338-for-peptide-level-evidence-under-the-rank-cut |
| co-IP of one candidate with MLL-AF4 in Jurkat | whether the association reproduces in the same system | low-moderate wet work |
| co-IP of one candidate with endogenous KMT2A::AFF1 in SEM or RS4;11 | whether it holds in the index disease and lineage | moderate wet work, and the experiment that actually matters |
| BioID/TurboID on endogenous KMT2A::AFF1 in a B-lineage line | the whole question, properly | high, and it is the standing gap this corpus has carried for many cycles |
The order matters. Running the deposit parse first is not merely cheapest — it decides whether any of the wet work is worth doing, because if the candidates rest on one or two ambiguous PSMs there is nothing to co-IP.
| absence class | measured-not-analysed |
|---|---|
| absence class note | The primary data is public in PXD009338 (100 files, 33.18 GB, including 12 deposited search-result files) and has been since 2018-05-25. Nothing here needs a new sample; the peptide-level questions need a parse and the biological question needs a co-IP. This is NOT never-measured. |
| what is absent | Any orthogonal validation - co-IP, reciprocal pulldown, proximity labelling, structure, genetic interaction - of any of the 110 proteins recovered under MLL-AF4 and no other bait in Skucha 2018. |
| closes with | exp-q180-reanalyse-pxd009338-for-peptide-level-evidence-under-the-rank-cut for the identification half; a co-IP in a B-lineage KMT2A-rearranged line for the biological half. |
| shown for fusion | The absence is ABOUT MLL-AF4's specific partner list, from a Jurkat T-ALL host with a tagged single-copy transgene. Nobody has looked at any of the 110 in a B-lineage KMT2A-rearranged cell by any method. |
| epigenetic readout | none for any of the 110. The source paper's own epigenetic finding - H3K36me3 and genome integrity - runs through SETD2, a pan-fusion partner. |
bounded by
[ "The source paper's own follow-up, which screened 40 genes drawn from the >=5-of-7 shared class and validated SETD2, a pan-fusion partner (inferred from the Fig. 2a funnel; the 40-gene list is not in the text)", "IntAct: PMID 29777171 never curated, per two existing corpus entries, not re-verified live", "PubMed titles and abstracts, 2026-09-24, one symbol per gene, name lists declared and incomplete", "This corpus: 89 of 110 never mentioned across 3,045 node files" ]
not bounded by
[ "BioGRID, STRING, MINT - not queried", "Supplementary figures of any paper - rule 80", "Alias and retired symbols for five of the six candidates - CCDC46, MCAD, GDH1, RPA32/RFA2, LEF-1, RANBP2L3", "The complex form 'RPA complex' - rule 64", "PRIDE or MassIVE swept for a second MLL-fusion interactome that might replicate any of the 110" ]
measurements
[
{
"quantity": "count",
"subject": "MLL-AF4-specific preys with any orthogonal validation",
"value": 0,
"units": null,
"n": 110,
"universe": "the 110 preys recovered under MLL-AF4 alone in Skucha 2018 Suppl Table 5, bounded by the source paper's own follow-up, IntAct curation status, PubMed tiab queries on six of them and a corpus-content join",
"null": null,
"note": "rule of three: with 0 of 110 validated and the bound as declared, the true rate is below 2.7% with 95% confidence - though the PubMed half of the bound covers only 6 of the 110 by name, so this interval is generous to the bound"
},
{
"quantity": "count",
"subject": "MLL-AF4-specific preys never mentioned in this corpus",
"value": 89,
"units": null,
"n": 110,
"universe": "word-boundary match over 3,045 node files excluding this beat's folder, 2026-09-24",
"null": 0,
"note": "1 has its own entry (KMT2D); 6 of the 21 mentioned appear only in the entry that reported the gap"
}
]- An absence of validation is not a refutation, and this entry must not be cited as one. It says nobody has looked, which is a statement about the field's attention and about the source paper's own priorities, not about whether the associations are real.
- The rule-of-three bound quoted here (true validation rate below 2.7%) is generous to itself: the PubMed half of the bound covers six of the 110 by name, so the other 104 are bounded only by IntAct curation status and this corpus's own contents. A reader should not treat the interval as if all 110 had been searched individually.
- Which 40 genes did Skucha 2018 screen? The list is not in the body text and would settle directly whether any AF4-specific partner was ever carried forward. It is one supplementary file away and was not opened.
- Does BioGRID hold PMID 29777171? One query, and it would change whether 'no curated record exists' is true.
- Is there a second MLL-fusion interactome anywhere in PRIDE or MassIVE? Neither archive was swept. A second dataset would let any of the 110 be tested by replication rather than by new experiment, and is the single highest-value unrun query in this beat.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Skucha 2018's functional follow-up was seeded from the shared end of its own interactome: 128 proteins interacting with at least five of seven fusions, from which 40 were screened, yielding SETD2 — and SETD2 is recovered under all seven baits (MLL-AF4 p = 0.0372 on 3|4 spectra), not in the MLL-AF4-specific set.
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 2018What this evidence cannot show
Read from the Europe PMC full-text XML of PMC5959866 on 2026-09-24. The paper states the 40 screened genes were drawn from the filtering funnel shown in Fig. 2a but does NOT list them, and the 128-protein figure is the >=5-of-7 class, so 'seeded from the shared end' is an inference from the funnel description rather than a statement the paper makes about its selection rule. Whether any of the 40 came from a singleton list cannot be determined without the gene list, which is not in the text and was not looked for in the supplementary files.
Evidence that argues against this (1)
- This is a genuine gap in the evidence for this entry: the 40 screened genes are not enumerated in the text, so the claim that NO AF4-specific partner entered the screen is not directly verified - it is inferred from the funnel figure's description. The 40 may be listed in a supplementary table this beat did not open. A reader should treat 'none of the 110 was screened' as probable rather than established until that list is checked, which is one supplementary file away. Nature Communications 2018
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
No curated interaction record exists for any of the 110: IntAct has never curated PMID 29777171, so the entire Skucha interactome — specific tail and shared core alike — sits outside every curated interaction database this project queries.
The index fusion has an AP-MS interactome and this graph never cited it; and IntAct holds seven MLL-fusion proteins and not the index one · This repository, entries ana-the-index-fusion-has-an-ap-ms-interactome-and-this-graph-never-cited-it and absence-intact-holds-seven-mll-fusion-proteins-and-not-the-index-one 2026What this evidence cannot show
Quoted from existing corpus entries and NOT re-verified against live IntAct by this beat. Rule 90 applies directly: IntAct gains depositions, and a figure drawn once from a live database and never refreshed goes stale invisibly. Anyone relying on this should re-run IntAct REST findInteractions for Q03164 and check the publication list for 29777171. Only IntAct was considered; BioGRID, STRING and MINT were not checked and BioGRID in particular curates supplementary tables that IntAct does not.
Evidence that argues against this (1)
- BioGRID curates high-throughput supplementary tables aggressively and may well hold this dataset even though IntAct does not, which would mean 'no curated record exists' is false. That check was not run, and it is a single query. The narrower claim that survives regardless is that no IntAct-derived coverage figure in this project can see these 110 proteins - which is the claim that matters for how the gap stayed invisible. BioGRID, thebiogrid.org 2026
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
None of the six p<0.05 MLL-AF4-specific candidates has any independent literature linking it to KMT2A or MLL: on 2026-09-24 RPA2 AND KMT2A = 0, RPA2 AND MLL = 0, LEF1 AND KMT2A = 1 (a bladder-cancer PRMT5-WDR5 paper), LEF1 AND MLL = 2 (both off-topic), CEP112 AND leukemia = 0, ACADM AND leukemia = 1, GLUD1 AND leukemia = 2, and RGPD8 returns 2 records in all of PubMed.
PubMed esearch via scripts/pubmed_safe.py, run 2026-09-24 by Q180. One [tiab] tag per term (rule 94); querytranslation read back and non-empty for every query, so these are genuine zeros and not instrument failures (rule 97). · NCBI PubMed E-utilities 2026What this evidence cannot show
Bounded by titles and abstracts only (rule 80) - an orthogonal validation reported in a supplementary figure is invisible to this, and for a co-IP that is a plausible place for it to live. The name lists are one symbol per gene and are weak: CCDC46 (CEP112), MCAD (ACADM), GDH1 (GLUD1), RPA32/RFA2 and the complex form 'RPA complex' (RPA2), LEF-1 and TCF1-alpha (LEF1), and RANBP2L3 (RGPD8) were not all tried. Rules 64, 71 and the retired-symbol rule all bite here. Every count is a floor.
Evidence that argues against this (1)
- Searching LEF-1 rather than LEF1 does change one of these numbers - it returns PMID 16510874, the beta-catenin/MLL H3K4 paper, which the LEF1 spelling misses. So the name-list weakness declared in the caveat is not hypothetical; it demonstrably altered one result in this very set. The counts should be read as establishing that no DEDICATED follow-up of any of the six has been published under its official symbol, not as establishing that the literature is empty. Sierra J et al. 2006
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
Within this corpus the 110 are almost entirely absent: 89 of 110 are never mentioned in any node file, 1 has its own gene entry (KMT2D, written for nomenclature reasons unrelated to this table), and 6 of the 21 that are mentioned appear only in the cycle-158 entry that reported the gap.
Mention audit of the 110 MLL-AF4-specific preys over 3,045 node files excluding this beat's own folder, run 2026-09-24 by Q180. Output analyses/q180-is-af4-specificity-a-depth-artefact/mentions_of_the_tail.json, which records the matching entry ids per symbol and not only the counts. · This repository (script and JSON committed; deterministic) 2026What this evidence cannot show
A word-boundary regex over serialised JSON counts a symbol appearing in any field, including a reference title or a caveat, and cannot distinguish substantive discussion from list membership - which is precisely why the 6 self-referential cases were separated out. The audit deliberately excludes this beat's own folder so the figure describes the corpus as it stood before this cycle; re-running it now will give larger numbers.
Evidence that argues against this (1)
- Absence from this corpus is a statement about this corpus and not about the world. The graph has 3,000-odd entries against a field of tens of thousands of papers, and 89 unmentioned proteins is unremarkable for any arbitrary 110-protein list. This evidence item is only meaningful because the list in question is nominally the index fusion's specific interactome, i.e. the PI's first stated priority - against any other reference set it would say nothing. This repository, entry ana-q178-the-interactome-gap-is-not-where-the-standing-priority-says-it-is 2026
nothing external has ever tested any of the 110, which is why the question is open rather than answered
the one partner the source paper did validate is SETD2, recovered under all seven baits - the opposite list
closes the identification half of this absence; the biological half needs a co-IP in a B-lineage line
the coverage inversion and the absence of any external validation are two views of the same fact: nobody, inside or outside this graph, has worked on the tail
the hypothesis that the question is open rests on nothing external having tested any of the 110