Oncofusion

start / graph / ana-q129-the-scaffold-as-bait-audit-q104-asked-for-kmt2a-stops-in-2015-and-psip1-in-2012

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q129-the-never-discussed-partners-of-kmt2a-and-psip1

The REST-wide direction audit Q104 asked for: scaffold-as-bait rows DO exist in the 36% of the record PSICQUIC hides - KMT2A is a bait in 96 of 376 rows across 18 publications and PSIP1 in 11 of 114 across 4, but the most recent is 2015 and 2012, and none of the thirteen never-discussed partners comes from one

also called Q104 open question 2, answered · scaffold-as-bait direction audit · KMT2A bait 2015 · PSIP1 bait 2012

Somebody checked earlier whether anyone had ever tagged the proteins this project studies and asked what sticks to them, and found nobody had - but only in a slice of the database. Re-running it over the whole database shows they have been tagged, just not since 2015 for one and 2012 for the other.

The same thing, for a scientist

**`ana-q104-every-row-in-the-read-queue-is-prey-side-and-seven-members-are-rejected` closed with an explicit request:** *"Re-run the direction audit against IntAct REST. PSICQUIC returns 651 of the 1,015 rows REST gives for the six scaffold accessions, so 36% of the record was never inspected, and a scaffold-as-bait row could be sitting in it - which would be the single most consequential correction to this entry."* **This is that re-run, for KMT2A and PSIP1.** All 376 IntAct REST records touching Q03164 and all 114 touching O75475 were pulled and their `experimentalRole` fields counted [E1]. **Q104's worry was justified and its conclusion survives, both at once.** Scaffold-as-bait rows do exist in the wider record: **KMT2A is the experimental bait in 96 of 376 rows across 18 publications, and PSIP1 in 11 of 114 across 4** [E1]. So the sentence *"nobody has ever tagged KMT2A or PSIP1"* is false at corpus scale and was never asserted at corpus scale - Q104 scoped it to the read queue, correctly. **What the re-run adds is a date.** The most recent publication in which KMT2A is a bait is **PMID 25751424, Nat Struct Mol Biol, April 2015**; for PSIP1 it is **PMID 22872640, J Biol Chem, October 2012** [E1, E2]. **And the direction finding holds exactly where it matters: 0 of the 13 never-discussed partners of KMT2A and PSIP1 comes from an experiment in which KMT2A or PSIP1 was the bait** - 8 are prey-side affinity or proximity rows and 5 are bait-free crosslinks [E3].

Go deeper
The measurement
KMT2A (Q03164)PSIP1 (O75475)
IntAct REST rows376114
rows where it is bait9611
rows where it is prey16591
rows where it is neutral component (crosslink)468
rows unspecified183
publications where it is bait184
publications where it is prey4029
most recent bait publicationPMID 25751424, Nat Struct Mol Biol, Apr 2015PMID 22872640, J Biol Chem, Oct 2012

The 18 KMT2A-bait publications are PMIDs 10656681, 11313484, 11976948, 15199122, 17913746, 18598942, 18840606, 20153263, 20541251, 20541477, 20581860, 20818375, 21113167, 22327296, 22722839, 23870121, 24735925 and 25751424. The four PSIP1-bait publications are 18598942, 22748127, 22773103 and 22872640. Every one is a focused small-scale study from the era before proteome-scale interactomics 12.

What this means for the direction finding. The two facts are not in tension. KMT2A and PSIP1 were used as baits, repeatedly, by the chromatin field in 2000-2015. What has not happened since is any bait-side experiment curated into IntAct - so every partner added to their interactomes in the last decade is a by-product of somebody else's bait, which is precisely the structural point Q104 made.

Applied to the thirteen partners this beat adjudicated
partnerbaitthis project's proteinmethodsource
ABT1, ANOS1, FGFBP1, KRR1, OASL, PRKRAthe partnerKMT2A = preyanti-tag co-IPBioPlex 3.0, PMID 33961781
IRF2the partnerKMT2A = preypull downHubel ISG network, PMID 30833792
UTP3the partnerPSIP1 = preyBioID2 proximityKhoroshkin, PMID 39251607
GDF5, IGHMBP2, PPP1R13B, TMEM214nonePSIP1 = neutral componentDSSO crosslinkFasci, PMID 30021884
TMCO1noneKMT2A = neutral componentDSSO crosslinkFasci, PMID 30021884

Zero of thirteen. Eight are prey-side; five come from a bait-free crosslinking dataset in which all 4,856 IntAct rows record both participants as neutral component 3.

This corrects a premise in the dispatch brief for this beat, which described the thirteen as "every IntAct-curated human partner... a single publication and a single bait". There is no bait behind any of them in the sense that matters: nobody set out to ask what binds KMT2A or PSIP1 and got these answers.

Two instrument notes, both of the silent-false-zero family

1. The IntAct REST endpoint's field-scoped publication query returns a clean zero. Measured live 2026-09-23:

findInteractions/pubid:39251607   ->  totalElements 0
findInteractions/39251607         ->  totalElements 19380
findInteractions/pubid:30021884   ->  totalElements 0
findInteractions/30021884         ->  totalElements 4856

Same shape as rule 85 (PUB_TYPE unquoted) and rule 95 (unquoted t(4;11)): a well-formed, error-free zero from a query that does not compose. pubid: is the correct PSICQUIC field name and it is not the IntAct REST field name. Anyone bounding a publication's contents on this endpoint with the field-scoped form will record a fabricated absence.

2. A 23-row difference between endpoints, recorded rather than harmonised. ana-rbp-bioid2-proximity-map-k562 records 19,403 rows for PMID 39251607 via PSICQUIC; IntAct REST on 2026-09-23 returns 19,380 after filtering on publicationPubmedIdentifier. Bait count (50), prey count (2,682), median (5), mean (7.23) and maximum (40) all reproduce exactly 4. Neither number is wrong; they are different endpoints on different dates. Per rule 93's fourth clause, an endpoint change is announced with both numbers rather than silently substituted.

Facts and figures
scriptad hoc, not scripted: IntAct REST findInteractions paginated to exhaustion for Q03164 (376 rows), O75475 (114 rows), 39251607 (19,380 rows) and 30021884 (4,856 rows), with experimentalRole, detectionMethod, expansionMethod, hostOrganism, intactMiscore and publicationPubmedIdentifier read per row. Raw pulls are in the session scratchpad as q129_rest_Q03164.json, q129_rest_O75475.json and q129_pub_*.json and are NOT persisted to the repository.
statusrun
result summaryKMT2A is bait in 96/376 rows across 18 publications (most recent PMID 25751424, April 2015); PSIP1 is bait in 11/114 across 4 (most recent PMID 22872640, October 2012); 0 of the 13 never-discussed partners comes from a scaffold-as-bait experiment.
answers open question ofana-q104-every-row-in-the-read-queue-is-prey-side-and-seven-members-are-rejected, open question 2
answer statusANSWERED for KMT2A and PSIP1; NOT RUN for MEN1, AFF1, AFF4 and DOT1L
absence classmeasured-not-analysed
absence class noteThe direction audit for MEN1, AFF1, AFF4 and DOT1L is not a missing experiment - the records are in IntAct and the pull takes four requests. It was out of scope for this beat and is the obvious next increment, which is why this is measured-not-analysed rather than never-measured.
inputs
[
  "IntAct REST findInteractions, 2026-09-23",
  "NCBI esummary for the bait publications, 2026-09-23",
  "analyses/core-complex-interactome-gap/results.json field union_never_discussed"
]
reproduces
{
  "ana-rbp-bioid2-proximity-map-k562": "50 baits, 2,682 preys, median 5, mean 7.23, max 40 - exact; row count 19,380 here against 19,403 there (REST vs PSICQUIC, different dates)"
}
instrument traps found
[
  "IntAct REST findInteractions returns a clean totalElements 0 for the field-scoped form pubid:<PMID>; the bare PMID as free text returns the real count. Same family as rules 85 and 95."
]
How this could bite us back
  • The practical consequence is a work-order one: if no scaffold-as-bait experiment has been curated for a decade, then every interactome-derived intervention hypothesis in this graph is built on other people's questions, and the single missing reagent is a tagged KMT2A or KMT2A-fusion bait in a leukaemia line.
  • This audit could be read as licence to discount every prey-side row, which would be wrong: WDR5, RBBP5 and ASH2L are also prey-side in BioPlex and they are the real MLL1 complex. Direction is a reason to ask for more, not a filter that can stand alone.
Still open
  1. Run the same REST direction audit for MEN1, AFF1, AFF4 and DOT1L. Four requests, and it completes the answer to Q104's open question rather than answering a third of it.
  2. Has any KMT2A-bait or PSIP1-bait interaction experiment been published since 2015 and 2012 respectively, and simply not curated by IntAct? This bounds the database and not the field, and a single PubMed search would tell the two apart - which matters, because 'the field stopped asking' and 'IntAct stopped curating' have completely different consequences.
  3. Of the 95 KMT2A and PSIP1 partners that DO have entries in this graph, how many rest on prey-side rows only? Q104 asked this and it is still unanswered; the answer decides whether 'this project has 95 documented interactors' needs the same qualification the never-discussed thirteen do.
  4. Should this project file a curation request with IntAct for the pubid: behaviour, or is it correct behaviour being queried in the wrong language? The trap is real for this project either way and belongs in data/symbol-collisions.json or an equivalent registry.
Every source behind this entry (4)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Across all 376 IntAct REST records touching KMT2A (Q03164), KMT2A is the experimental bait in 96 rows spanning 18 publications, prey in 165 rows spanning 40 publications, neutral component in 46 and unspecified in 18. Across all 114 records touching PSIP1 (O75475), PSIP1 is bait in 11 rows spanning 4 publications (PMIDs 18598942, 22748127, 22773103, 22872640), prey in 91 rows spanning 29 publications, neutral component in 8 and unspecified in 3. The 18 KMT2A-bait publications are PMIDs 10656681, 11313484, 11976948, 15199122, 17913746, 18598942, 18840606, 20153263, 20541251, 20541477, 20581860, 20818375, 21113167, 22327296, 22722839, 23870121, 24735925 and 25751424.

    What this evidence cannot show

    This measures IntAct's curation, NOT the literature. A KMT2A-bait or PSIP1-bait experiment published after 2015 or 2012 and not curated into IntAct is invisible to this count, and no PubMed search was run to check - so the dates bound the database and nothing else. Only two of the six scaffold accessions in Q104's audit were re-run here; MEN1, AFF1, AFF4 and DOT1L were not, so Q104's open question is closed for a third of its scope. A row where KMT2A appears as a processed chain (Q03164-PRO_0000390949 or -PRO_0000390950) is counted under Q03164 by the REST endpoint, which is the behaviour rule 93 recommends but which means the 376 is not a count of full-length-protein rows.

    Evidence that argues against this (1)
    • The bait/prey annotation is IntAct's curation of the source experiment, not a field the experimenters filled in. For reciprocal co-immunoprecipitations and for gel-filtration or complex-purification experiments the assignment can be arbitrary, and 46 of KMT2A's rows carry no role at all. So the 96/165 split should be read as a strong signal about the shape of the record, not as an exact census of who tagged what. IntAct (EMBL-EBI) 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The most recent IntAct-curated publication in which KMT2A is an experimental bait is PMID 25751424, 'NAD(+)-SIRT1 control of H3K4 trimethylation through circadian deacetylation of MLL1', Nature Structural & Molecular Biology, April 2015. The most recent in which PSIP1 is a bait is PMID 22872640, 'Interaction of the HIV-1 intasome with transportin 3 protein (TNPO3 or TRN-SR2)', Journal of Biological Chemistry, October 2012.

    What this evidence cannot show

    Titles, journals and dates were read from a live esummary fetch in this session, not from memory or from a sibling entry. The claim is about which publications IntAct has curated, so a more recent bait experiment that IntAct has not reached would make the dates wrong as a statement about the field; they are correct as a statement about the database. Only the most recent PMIDs in each list were date-checked individually; the ordering of the remainder was inferred from PMID magnitude, which is reliable but not a fetch.

    Evidence that argues against this (1)
    • Both proteins have been intensively studied since those dates. The field moved to ChIP-seq, cryo-EM, degron systems and CRISPR screens, and IntAct curates interaction experiments rather than occupancy or structural work, so the drop-off plausibly reflects a change in what experiments get done and curated rather than a loss of interest. Reading these dates as 'nobody has pulled on KMT2A since 2015' would be a false absence of exactly the kind rule 93 warns about. IntAct (EMBL-EBI) 2026
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Of the thirteen IntAct-curated human partners of KMT2A or PSIP1 never discussed anywhere in this corpus, zero come from an experiment in which KMT2A or PSIP1 was the bait. Eight are prey-side spoke-expanded rows in which the partner was the tagged bait: ABT1, ANOS1, FGFBP1, KRR1, OASL and PRKRA from BioPlex 3.0 (PMID 33961781, anti-tag co-IP), IRF2 from PMID 30833792 (pull down), and UTP3 from PMID 39251607 (BioID2 proximity). Five come from PMID 30021884, a DSSO crosslinking dataset in which all 4,856 IntAct rows record both participants as 'neutral component' with no bait and no spoke expansion: GDF5, IGHMBP2, PPP1R13B and TMEM214 on PSIP1, and TMCO1 on KMT2A.

    What this evidence cannot show

    The thirteen-partner list was taken from analyses/core-complex-interactome-gap/results.json field union_never_discussed and independently verified by word-boundary grep of all 13 symbols plus 9 common aliases across every file in data/graph/nodes - all returned zero matches, case-insensitively. That list is itself derived from PSICQUIC and is therefore a floor: a partner visible only through REST would not be on it, so 'thirteen' is the number of never-discussed partners the instrument can see, not necessarily the number that exist.

    Evidence that argues against this (1)
    • A crosslink with no bait is not the same failure as a prey-side affinity row, and lumping the five Fasci rows with the eight affinity rows under 'nobody used our protein as bait' flattens a real methodological difference. A bait-free method cannot have a direction, so counting it as a direction failure is a category slip; the five are declined on other grounds in their own entries. Fasci D et al. 2018
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The IntAct REST findInteractions endpoint returns totalElements 0 for the field-scoped forms 'pubid:39251607' and 'pubid:30021884', and 19,380 and 4,856 respectively for the bare PMIDs as free text. The bare-PMID pull for 39251607 resolves to 50 distinct baits and 2,682 distinct preys with median 5 baits per prey, mean 7.23 and maximum 40 - reproducing the bait count, prey count, median, mean and maximum recorded in ana-rbp-bioid2-proximity-map-k562 exactly, against that entry's PSICQUIC row count of 19,403.

    What this evidence cannot show

    The bare-PMID form is free text and may match the number in fields other than the publication identifier, so it is an upper bound; every returned row was filtered on publicationPubmedIdentifier before counting and all of them passed for both publications, so no inflation was observed here. This was tested on two PMIDs only. The correct field-scoped syntax for this endpoint was not determined - 'pubid:' merely fails - so this note records a trap without supplying the fix.

    Evidence that argues against this (1)
    • The IntAct REST service has its own documented query syntax and 'pubid:' is a PSICQUIC MIQL field. Calling this a false zero is arguably unfair to the endpoint: it is being handed a query in another service's language. The reason it belongs in this project's registry anyway is that PSICQUIC and REST are used interchangeably here, the field name transfers silently, and the failure returns a number rather than an error. IntAct (EMBL-EBI) 2026
Connected to