Oncofusion

start / graph / ana-menin-is-the-best-observed-of-the-six-core-proteins

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q76-spoke-provenance

Menin is the best-observed of the six core proteins and the only one whose interval separates from the rest - the interface the programme is built on is the one the record actually supports

also called MEN1 46.7% spoke-only · the menin interactome is directly observed · spoke-expansion by core protein

Across the six proteins at the centre of this project, most recorded partnerships were never tested one-on-one. Menin is the exception: fewer than half of its partnerships are of that inferred kind, and it has eight partners whose contact has been measured directly - more than the other five put together on that measure.

The same thing, for a scientist

Grading every IntAct partner pair of the six core proteins by whether it rests only on MI:1060 spoke-expanded rows [E1]: MEN1 46.7% (60 pairs, Wilson 95% CI 34.6-59.1%), KMT2A 71.9% (146, 64.1-78.6%), PSIP1 73.2% (71, 61.9-82.1%), AFF4 75.9% (87, 65.9-83.6%), AFF1 76.7% (43, 62.3-86.8%), DOT1L 78.6% (42, 64.1-88.3%). **MEN1's interval does not overlap AFF1's, AFF4's or DOT1L's.** On the stricter measure - partners with any MI:0407 direct-interaction record - MEN1 has eight (JUND, KMT2A, KMT2B, MYH9, NFKB1, NFKB2, NME1, RELA) against AFF1's one and DOT1L's one. The load-bearing pair itself is the strongest in the set: **MEN1-KMT2A is 15 records from four publications, 5 spoke-expanded and 10 not, with MI:0407 direct interaction detected by ITC, molecular sieving, pull down and x-ray diffraction, and seven PDB entries containing both chains** - 3U85, 3U88, 4GQ6, 9C4S, 9C4T, 9C4U, 9C4V [E1][E2][E3]. This is a *supporting* result for the menin-interface hypothesis and it was not the expected one: the census was designed to find load-bearing claims resting on spokes, and the most load-bearing claim in the project turned out to be the best-evidenced pair in the whole set.

Go deeper
The ranking, with intervals
core proteinIntAct pairsspoke-only95% CIpartners with any MI:0407 record
MEN16046.7%34.6-59.1%8
KMT2A14671.9%64.1-78.6%10
PSIP17173.2%61.9-82.1%3
AFF48775.9%65.9-83.6%3
AFF14376.7%62.3-86.8%1
DOT1L4278.6%64.1-88.3%1

The five non-menin proteins are indistinguishable from each other - every interval overlaps every other - and reporting any of them against any other would be reporting noise. The separation that carries information is menin against the rest, and even that is a single comparison: MEN1's upper bound 59.1% clears AFF1's lower bound 62.3%, AFF4's 65.9% and DOT1L's 64.1%, but not KMT2A's 64.1% or PSIP1's 61.9%, which it approaches without clearing.

So the defensible statement is: menin is better observed than AFF1, AFF4 and DOT1L; against KMT2A and PSIP1 this sample cannot resolve it. That is narrower than "menin is the best-observed" and it is what the data support.

KMT2A's ten direct-interaction partners include two histone entries (H3C12 and an H3.2 record) and a mouse Kat8, so the like-for-like human protein-partner count is lower than the raw ten.

The menin-KMT2A pair itself
IntAct records15, across four publications (PMID 15199122, 18598942, 20818375, 22327296)
spoke-expanded5
not spoke-expanded10
strongest typeMI:0407 direct interaction
detection methodsanti-bait co-IP, anti-tag co-IP, ITC, molecular sieving, pull down, x-ray diffraction
PDB entries containing both chains3U85, 3U88, 4GQ6, 9C4S, 9C4T, 9C4U, 9C4V 3

This is the only pair in the census supported by an isothermal-titration-calorimetry record, a size-exclusion record, a crystallographic record and a co-immunoprecipitation record at once. Whatever else the census found, it did not find a weakness here.

One record deserves naming because of how nearly it was missed: EBI-15969392, KMT2A-MEN1, MI:0407 direct interaction, PMID 22327296 is filed with menin as isoform O00255-2, and is therefore invisible to the PSICQUIC id:O00255 query that this project's earlier interactome analyses use 4. The single most important interaction record in the domain was outside the instrument.

The one soft spot inside the menin arm: menin-LEDGF

MEN1-PSIP1 is one IntAct record, anti-bait co-IP, spoke-expanded, MI:0914 association, from PMID 18598942 5. It is the weakest-supported pair in the menin arm and it sits inside the three-body argument the project uses most.

The pair does share PDB 3U88, "Crystal structure of human menin in complex with MLL1 and LEDGF", whose three entities are O00255, Q03164 and O75475 3. But a shared entry is co-membership of one observed assembly, not a contact - and in this particular assembly the well-described contacts are menin-MLL (MBM1/MBM2 into the menin pocket) and LEDGF-MLL (IBD onto the MLL IBM). Whether menin and LEDGF touch each other directly in 3U88 was not resolved in this beat and should not be asserted from a cross-reference intersection.

That is not a problem for the ternary model, which never required a direct menin-LEDGF contact - MLL bridges them by design. It is a caution against reading the PDB-intersection column as though it graded contacts, and it is recorded in the census entry's caveats for exactly that reason.

What would make this result go away
  1. Curation effort, not biology. Menin is a MEN1-syndrome tumour suppressor with a thirty-year clinical literature and a drug programme; it has been studied by structural biologists because it is druggable. The same fact that makes it the target makes it the best-curated. A protein is directly observed because people looked, and people look at proteins that are already interesting - so this measure cannot distinguish a genuinely better-evidenced interface from a better-funded one.
  2. Denominator asymmetry. MEN1 has 60 pairs against KMT2A's 146. A smaller, more selectively curated partner list will have a higher directly-observed fraction by construction, because the marginal partner added by an atlas is always a spoke.
  3. It is one comparison in a six-way table. No multiplicity correction is applied and none is claimed; the finding is presented as a ranking with intervals, not as a test.
Facts and figures
scriptanalyses/spoke-expansion-provenance-census/run.py
statusrun
result summaryMEN1 is 46.7% spoke-only against 71.9-78.6% for the other five, with an interval that separates from AFF1, AFF4 and DOT1L but not from KMT2A or PSIP1; MEN1-KMT2A is the best-evidenced pair in the census (15 records, 10 not spoke-expanded, ITC + x-ray + SEC + co-IP, seven shared PDB entries); MEN1-PSIP1 is the weakest pair in the menin arm at one spoke-expanded row.
per core
DOT1L42.0DOT1L: 42AFF143.0AFF1: 43MEN160.0MEN1: 60PSIP171.0PSIP1: 71AFF487.0AFF4: 87KMT2A146.0KMT2A: 146
inputs
[
  "IntAct REST ws/interaction/findInteractions for the six core accessions, 2026-09-21",
  "UniProt PDB cross-references for O00255, Q03164, O75475",
  "RCSB PDB entry and polymer-entity records for 3U88"
]
per core — the numbers behind the chart
{
  "MEN1": {
    "pairs": 60,
    "pct_spoke_only": 46.7,
    "wilson_95": [
      34.6,
      59.1
    ],
    "direct_partners": [
      "JUND",
      "KMT2A",
      "KMT2B",
      "MYH9",
      "NFKB1",
      "NFKB2",
      "NME1",
      "RELA"
    ]
  },
  "KMT2A": {
    "pairs": 146,
    "pct_spoke_only": 71.9,
    "wilson_95": [
      64.1,
      78.6
    ]
  },
  "PSIP1": {
    "pairs": 71,
    "pct_spoke_only": 73.2,
    "wilson_95": [
      61.9,
      82.1
    ]
  },
  "AFF4": {
    "pairs": 87,
    "pct_spoke_only": 75.9,
    "wilson_95": [
      65.9,
      83.6
    ]
  },
  "AFF1": {
    "pairs": 43,
    "pct_spoke_only": 76.7,
    "wilson_95": [
      62.3,
      86.8
    ]
  },
  "DOT1L": {
    "pairs": 42,
    "pct_spoke_only": 78.6,
    "wilson_95": [
      64.1,
      88.3
    ]
  }
}
men1 kmt2a
{
  "n_records": 15,
  "n_spoke_expanded": 5,
  "pmids": [
    "15199122",
    "18598942",
    "20818375",
    "22327296"
  ],
  "detection_methods": [
    "anti bait coip",
    "anti tag coip",
    "itc",
    "molecular sieving",
    "pull down",
    "x-ray diffraction"
  ],
  "shared_pdb_entries": [
    "3U85",
    "3U88",
    "4GQ6",
    "9C4S",
    "9C4T",
    "9C4U",
    "9C4V"
  ],
  "record_invisible_to_psicquic": "EBI-15969392 (menin filed as O00255-2)"
}
men1 psip1
{
  "n_records": 1,
  "n_spoke_expanded": 1,
  "interaction_ac": "EBI-1801742",
  "pmids": [
    "18598942"
  ],
  "shared_pdb_entries": [
    "3U88"
  ],
  "note": "shared entry is co-membership of the ternary complex, not an established menin-LEDGF contact"
}
How this could bite us back
  • This is the pleasant result and it should be held to a higher bar than an unpleasant one. The census was designed to find a load-bearing claim resting on a spoke; it did not find one at the menin interface, and the correct reaction to a designed test failing to fire is suspicion of the test, not relief.
  • Even a perfectly observed menin-MLL contact says nothing about whether occupying it is therapeutically sufficient. The same pocket binds JunD with opposite transcriptional consequences, menin inhibitors also degrade menin, and every pivotal pharmacodynamic readout in the class is chromatin occupancy - which cannot separate eviction from degradation. Evidence quality for the interaction and mechanistic clarity for the drug are different quantities.
  • Ranking core proteins by observation quality invites a programme to concentrate on the best-observed one, which is already the one everybody is working on. The proteins this census grades worst - AFF1 above all - are the ones where a single well-designed binary experiment would move the most, and a quality ranking read as a priority ranking inverts that.
Still open
  1. Do menin and LEDGF make a direct contact in 3U88, or does MLL bridge them entirely? The coordinates would settle it in an afternoon, and the answer determines whether menin-LEDGF is a two-body interaction at all or only a co-membership.
  2. Is menin's advantage a property of its interactions or of how much they have been studied? A pairing of each core protein's partner pairs against citation counts for those pairs would separate the two, and the data for it are already in the census output.
  3. The four new menin-KMT2A PDB entries 9C4S-9C4V are recent depositions. What are they, and does any of them place a clinical menin inhibitor in the pocket with the MLL peptide still bound? This beat read them only as a cross-reference intersection.
  4. KMT2A's direct-interaction partner list includes two histone entries and a mouse Kat8 record. What is the like-for-like human protein-partner count, and does KMT2A's interval move enough to separate from menin once non-protein and non-human records are excluded?
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    In the IntAct REST record fetched 2026-09-21, the fraction of partner pairs resting only on MI:1060 spoke-expanded rows is MEN1 28/60 = 46.7%, KMT2A 105/146 = 71.9%, PSIP1 52/71 = 73.2%, AFF4 66/87 = 75.9%, AFF1 33/43 = 76.7% and DOT1L 33/42 = 78.6%; MEN1 has eight partners with at least one MI:0407 direct-interaction record (JUND, KMT2A, KMT2B, MYH9, NFKB1, NFKB2, NME1, RELA) against one each for AFF1 and DOT1L; and MEN1-KMT2A is 15 records of which 5 are spoke-expanded, with direct-interaction records detected by ITC and x-ray diffraction.

    What this evidence cannot show

    Wilson 95% intervals are reported for every proportion and five of the six overlap each other, so only the MEN1-versus-{AFF1, AFF4, DOT1L} separations are claimed. Six comparisons, no multiplicity correction, and none is claimed - this is a ranking with intervals, not a test. The unit is a pair and not an experiment. Curation effort is a confound that this design cannot remove and the entry says so in its own section. The counts are from the REST route; the PSICQUIC route used by this project's older interactome analyses reaches only 43.6% of MEN1's record and 51.4% of KMT2A's, so these denominators are not comparable to the ones in those files.

    Evidence that argues against this (1)
    • A directly observed interaction is one somebody chose to measure directly, and menin has been chosen more often than any of the other five because a drug programme depends on it. The ranking may therefore measure twenty years of pharmaceutical attention rather than any property of the interactions, and the two are not separable from a database count. The honest reading is that menin's interface is well supported AND that this measure would look the same if it were merely well funded. EMBL-EBI IntAct 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    PMID 22327296 (Huang J, Gurung B, Wan B, et al., Lei M, Nature 2012) supplies the menin-KMT2A direct-interaction records in IntAct, including the x-ray-diffraction record and the ITC record.

    PMID 22327296DOI
    What this evidence cannot show

    Title, nine-author list, journal, year and DOI copied from a live NCBI efetch XML record fetched 2026-09-21 with the anchored PubmedData/ArticleIdList/ArticleId path (brief rule 81). Already cited in 52 entries of this graph (brief rule 82) and its substantive findings about the shared pocket are written elsewhere; used here only to identify the experimental source of specific IntAct records. The mapping from PMID to record comes from each record's publication identifier, not from the paper's text.

    Evidence that argues against this (1)
    • This paper's headline is that the SAME menin pocket binds MLL and JunD with opposite transcriptional consequences, which means the pocket is not MLL-specific. A programme that occupies it is not occupying an MLL-dedicated site, and the strength of the menin-MLL evidence does not by itself make pocket occupancy a clean intervention. Huang J et al. 2012
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Seven PDB entries contain polymer entities mapped to both O00255 and Q03164 - 3U85, 3U88, 4GQ6, 9C4S, 9C4T, 9C4U and 9C4V - and 3U88 'Crystal structure of human menin in complex with MLL1 and LEDGF' additionally contains a third entity mapped to O75475.

    What this evidence cannot show

    The intersection is computed from UniProt cross-reference lists; only 3U88 was verified entity-by-entity against the RCSB polymer_entity endpoint, and the other six are asserted at the level of the cross-reference. A shared entry establishes that both proteins are in one deposited assembly and NOT that they are in contact - which matters here, because 3U88's shared entry with PSIP1 does not establish a menin-LEDGF contact and is not used as though it did.

    Evidence that argues against this (1)
    • Several of these entries are the same complex re-solved with different inhibitors bound, so seven entries is not seven independent observations. Counting depositions overstates independence in exactly the way counting IntAct rows does, which is a defect this census criticises elsewhere and should not commit here. UniProtKB 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    IntAct record EBI-15969392, KMT2A-MEN1, MI:0407 direct interaction from PMID 22327296, is filed with menin as isoform accession O00255-2 and is therefore not returned by a PSICQUIC id:O00255 query, which reaches 58 of MEN1's 133 interaction records.

    What this evidence cannot show

    Established by set-differencing interaction accessions between the two routes on the same day, and by reading the identifier on the record. Query-form controls for the PSICQUIC side are printed in the analysis output. The fuller treatment is in ana-psicquic-id-queries-miss-half-of-kmt2as-intact-record; this evidence item exists so the menin conclusion does not depend on reading that entry.

    Evidence that argues against this (1)
    • The menin-KMT2A pair was never actually going to be missed in practice, because it has fourteen other records and several are reachable through the narrow query. The invisibility of this one record is an instrument finding, not a claim that the project ever got the menin interaction wrong. EMBL-EBI IntAct PSICQUIC 2026
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    MEN1-PSIP1 is supported by exactly one IntAct record, EBI-1801742, anti-bait co-immunoprecipitation, MI:0914 association, spoke-expanded, from PMID 18598942.

    What this evidence cannot show

    Graded moderate rather than strong because it is a statement about a single curated row and the source publication (PMID 18598942) was NOT read in this beat - only its identifier on the record was. The menin-LEDGF relationship is extensively described in this graph's ledgf-axis entries on evidence this item does not touch, and nothing here should be read as bounding that literature; the claim is narrowly about what IntAct holds.

    Evidence that argues against this (1)
    • One curated row is a statement about curation depth. The menin-LEDGF-MLL ternary model rests on a published crystal structure and on functional work in this graph's ledgf-axis folder, none of which is affected by how many IntAct rows the binary pair has; treating the single row as the evidence base would badly understate the case. RCSB Protein Data Bank 2026
Connected to