Nobody has ever run proximity labelling on a KMT2A fusion, and the interaction records that look like they did are RBP-atlas preys from HEK293
also called BioID KMT2A fusion · TurboID proximity labelling · interactome provenance · prey versus bait
There is a standard modern method for mapping everything that comes near a protein inside a living cell, and it has never been applied to a KMT2A fusion. The partner database looks as though it has been, but those entries come from a study of unrelated proteins in which our protein merely happened to get tagged.
The same thing, for a scientist
**Bounded absence, controlled (rule 67).** A PubMed query pairing the proximity-labelling vocabulary — `proximity labeling/labelling, BioID, TurboID, APEX2, proximity-dependent biotin, miniTurbo` — with `KMT2A OR MLL OR mixed lineage leukemia` returns **0**. Controls: the method vocabulary alone **2,669**; KMT2A/MLL interactome or interaction proteomics in leukaemia **56**; the method plus leukaemia fusion/oncoprotein **14**, none on KMT2A. **No proximity-labelling experiment has been reported on any KMT2A fusion protein.** **And the apparent counter-evidence dissolves on opening it.** This project's core-complex interactome gap analysis pulled IntAct partner rows whose method is *proximity-dependent biotin identification*, for baits including KMT2A, MEN1 and PSIP1 — **BUD13, ILF3, RPS11 and YWHAG** among them. Those rows trace to **PMID 39251607**, which is *Systematic identification of post-transcriptional regulatory modules*: a **50-RNA-binding-protein proximity-labelling atlas in K562 cells**. KMT2A is not its subject, is not in its title or abstract, and appears only as **prey** detected near someone else's RBP bait.
The provenance trap, which is the transferable part
An interaction database row tells you a method and a publication. It does not tell you whose experiment it was. When the source is a large-scale atlas, your protein may be prey rather than bait — detected incidentally, in the atlas's cell line, under the atlas's conditions, with no one having thought about your biology at all.
Three consequences here:
- The literature query and the database disagree, and both are right.
proximity labelling AND KMT2Areturns 0 because no paper is about that; IntAct returns rows because KMT2A was in somebody's dataset. Querying only one of them gives the wrong answer in opposite directions. - The system is wrong for this graph. The baits are RNA-binding proteins and the cells are K562, a BCR::ABL1 CML line - a human leukaemia cell, but NOT a KMT2A-rearranged one. Rule 73 still applies, far less forcefully than if this had been a non-haematopoietic line.
- It downgrades part of a queue I wrote. The curated interactome read queue recorded in the coverage ledger last cycle ranks candidates partly by publication count. BUD13, ILF3 and RPS11 enter it on this atlas alone, and YWHAG on this atlas plus one other. Their '1 publication, proximity-dependent biotin identification' grade is accurate and misleading at once: it is one paper that was not studying KMT2A.
The rule that follows: before writing an entry from an interaction-database row, open the source and establish whether your protein was the BAIT. Prey-only evidence from a large-scale atlas is a lead at best, and its cell system is the atlas's, not yours.
The experiment this makes obvious, and why it is the PI's priority in one assay
Proximity labelling of KMT2A::AFF1 in an index-fusion line has never been done, and it is the single most direct instrument for the standing instruction to map every interactor.
Design, briefly: express a TurboID- or miniTurbo-tagged fusion at near-endogenous level in a KMT2A::AFF1 line (SEM or RS4;11), with two controls that matter more than the experiment — the tag alone, and tagged wild-type KMT2A, because the whole question is what the fusion recruits that the wild-type protein does not. Short biotin pulses to keep the labelling radius tight; streptavidin capture; mass spectrometry.
Why it is worth more here than another co-immunoprecipitation. This graph's interactome rests heavily on anti-tag co-IP, which reports stable complexes that survive lysis. Proximity labelling reports neighbourhood in a living cell, including transient and low-affinity contacts and anything held together by chromatin rather than by protein-protein affinity — which is precisely the regime the fusion is proposed to work in.
What would make it fail. Tagging a large chromatin protein can disrupt the complex it is meant to report; over-expression creates neighbours that do not exist at endogenous level; and the wild-type-KMT2A control is essential and hard, because the fusion and the wild-type allele coexist in these cells.
CORRECTED the same hour: the cells are K562, not HEK293, and I had it the flattering way round
Two checks were named in this entry's own caveats. Both were run and they went opposite ways.
Confirmed, and now stronger: KMT2A appears zero times in the full text of PMID 39251607. It is not a bait, and that no longer rests on absence from the abstract.
Wrong, and wrong in my favour: I wrote that the atlas was in HEK293, taken from the paper's MeSH terms. The full text says "we fused the BioID2 system to 50 RBPs in K562 cells". K562 is a human chronic myeloid leukaemia line, not embryonic kidney.
That materially weakens the argument built on it. 'The system is wrong for this graph' was the strongest-sounding of the three consequences, and a prey detection in a human leukaemia line is far more relevant here than one in kidney cells. K562 is BCR::ABL1 and not KMT2A-rearranged, so rule 73 still bites - but nothing like as hard as claimed.
The provenance point survives intact: these are prey detections from an RBP-directed atlas, not experiments on KMT2A. What does not survive is the characterisation of the cells - taken from a metadata field instead of the methods sentence, which is the same shortcut this project keeps getting caught by, and this time it inflated the conclusion.
bounded query
{
"target_hits": 0,
"controls": {
"method vocabulary alone": 2669,
"KMT2A interactome in leukaemia": 56,
"method + leukaemia fusion/oncoprotein": 14
},
"run": "PubMed esearch 2026-09-20"
}atlas source
{
"pmid": "39251607",
"n_baits": 50,
"bait_class": "RNA-binding proteins",
"cells": "K562 (BCR::ABL1 CML)",
"queue_entries_affected": [
"BUD13",
"ILF3",
"RPS11",
"YWHAG"
],
"title": "Systematic identification of post-transcriptional regulatory modules.",
"authors": "Khoroshkin M, Buyan A, Dodel M, et al.",
"venue": "Nat Commun",
"year": 2024,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}- This entry proposes an experiment and the honest reason it has not been done may be that it is hard rather than that nobody thought of it: tagging a large chromatin protein can break the complex it is meant to report, and the wild-type-KMT2A control is essential and awkward because both alleles coexist. Do not read 'nobody has done it' as 'it is easy and overlooked'.
- And do not strip BUD13, ILF3 and RPS11 out of the interactome queue on the strength of this. Prey-only detection in the wrong cell type is a weak lead, not a refutation, and this project has previously over-corrected after discovering a provenance problem.
- This entry got the cell line wrong in its first version - HEK293 instead of K562 - by reading a metadata field rather than the methods sentence, and the error ran in the direction that made its own argument stronger. Corrected within the hour. Treat the 'wrong system' half as weak and the bait-versus-prey half as the part that holds.
- ANSWERED 2026-09-20: KMT2A appears zero times in the full text of PMID 39251607, so it is not among the 50 baits. The bait table itself was not downloaded, but absence from the entire text is stronger than absence from the abstract.
- How many other rows in this graph's interactome queue are prey-only detections from large-scale atlases rather than experiments on these proteins? The bait/prey distinction was not recorded when the queue was built.
- Would a TurboID of KMT2A::AFF1 against a tagged wild-type KMT2A control actually separate fusion-specific neighbours, given that both alleles are present in the same cell?
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
No proximity-labelling experiment on any KMT2A fusion protein is retrievable from PubMed: a query pairing six proximity-labelling method terms with KMT2A, MLL or mixed lineage leukemia returns 0 records, against controls of 2,669 for the method vocabulary alone, 56 for KMT2A interactome or interaction proteomics in leukaemia, and 14 for the method with leukaemia fusion or oncoprotein terms, none of which concerns KMT2A.
Bounded absence query with controls for proximity labelling of KMT2A fusions, PubMed esearch 2026-09-20 · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
**Moderate because this is a negative retrieval claim**, and this entry itself demonstrates the failure mode: a large-scale atlas containing KMT2A as prey is invisible to a query pairing the method with the gene, because the gene is in neither title nor abstract. So the query bounds what has been PUBLISHED ABOUT KMT2A by this method, not what data exist. PubMed only - preprints, theses and supplementary datasets of papers about something else are not covered, and the latter is exactly where such data would hide.
Evidence that argues against this (1)
- IntAct does return proximity-dependent biotin identification rows for KMT2A, MEN1 and PSIP1, so data of this type involving these proteins demonstrably exist. They are prey detections in a 50-RBP atlas in HEK293 rather than experiments on KMT2A - which is why the claim here is about no experiment having been RUN ON a KMT2A fusion, not about no such rows existing. EMBL-EBI IntAct 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.journal-article
The IntAct proximity-dependent biotin identification rows that place BUD13, ILF3 and RPS11 near KMT2A, MEN1 or PSIP1 derive from a study whose subject is a 50-RNA-binding-protein proximity-labelling atlas in K562 cells, in which KMT2A is neither a bait nor named in the title or abstract, so those rows are incidental prey detections in a CML line rather than evidence about the leukaemic complex.
Systematic identification of post-transcriptional regulatory modules. · Khoroshkin M, Buyan A, Dodel M, et al. · Nat Commun 2024What this evidence cannot show
**CORRECTED 2026-09-20 from the full text (PMC11385195).** (1) **The cells are K562, not HEK293** - the full text says the BioID2 system was fused to 50 RBPs in K562 cells; HEK293 was taken from the MeSH terms. K562 is a human CML line, so these prey detections are considerably more relevant to a leukaemia graph than this entry first claimed, though K562 is BCR::ABL1 and not KMT2A-rearranged (rule 73). (2) **KMT2A appears zero times in the full text**, so 'not a bait' is now read rather than inferred. The bait table itself was still not downloaded.
Evidence that argues against this (2)
- Prey-only detection is not worthless. A protein repeatedly detected near many unrelated baits may be genuinely abundant or genuinely promiscuous, and BUD13, ILF3 and RPS11 are RNA-associated proteins in a graph whose index fusion sits in an elongation complex with real RNA biology. Dismissing the rows entirely would be the opposite error to over-reading them; the correct status is a lead whose system is wrong. EMBL-EBI IntAct 2026
- AND THE SYSTEM OBJECTION IS WEAKER THAN THIS ENTRY FIRST MADE IT. The atlas ran in K562, a human leukaemia line, not a non-haematopoietic cell. A KMT2A prey detection there is a lead in roughly the right tissue context, which removes most of the force of the 'wrong system' argument while leaving the bait-versus-prey argument untouched. Khoroshkin M et al. 2024
QUANTIFIES ITS POINT ACROSS THE WHOLE COMPLEX: 67.9% of partner-bait pairs are prey-only, and all seven top read-queue rows are.
Does NOT refute it — AP-MS is not proximity labelling and that absence survives — but reframes it. The entry's own control of 56 records for KMT2A/MLL interaction proteomics in leukaemia contains PMID 29777171, so the control number was reported and the records behind it were never opened.
WHY THAT EXPERIMENT MATTERS MORE THAN ANOTHER READING PASS: the constraint is that almost nobody has fished with these proteins, not that this graph has failed to transcribe them.
- related to →KMT2A / MLL1 — the gene that breaks
THE MOST DIRECT INSTRUMENT FOR THE STANDING PRIORITY, never applied to the fusion: no proximity-labelling experiment on any KMT2A fusion is retrievable.
ITS IntAct PROXIMITY ROWS ARE PREY DETECTIONS in a 50-RBP atlas in HEK293, not experiments on menin.
SAME PROVENANCE ISSUE: its proximity-labelling partner rows come from the RBP atlas.
A BOUNDED ABSENCE WHOSE OWN CAVEAT NAMES HOW IT COULD BE WRONG: a large-scale atlas is invisible to a method-plus-gene query.
THE SYSTEM PROBLEM IN A NEW PLACE: partner rows carrying no indication that the cells were embryonic kidney.