Oncofusion

start / graph / ana-q104-every-row-in-the-read-queue-is-prey-side-and-seven-members-are-rejected

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q104-interactome-read-queue

In all 24 read-queue rows inspected, the candidate is the BAIT and this project's protein is the prey - nobody has ever tagged KMT2A, AFF4, menin or PSIP1 and found any of them; plus seven bounded rejections

also called Q104 read-queue direction audit · Q104 bounded rejections · the read queue is entirely prey-side

Fifteen proteins are queued as unwritten interaction partners of the leukaemia machinery. Reading the raw records one by one shows that in every single case somebody was studying the other protein and ours happened to turn up - no experiment has ever started from our proteins and found any of these - and seven of the fifteen should not get entries at all, for reasons recorded here with their numbers.

The same thing, for a scientist

**The single measurement of this beat.** The MITAB2.7 experimental-role column was read for every row in the read queue rather than assumed. Of the **24 rows** covering the thirteen non-substrate queue members this beat inspected directly, **24 have the candidate as experimental bait (MI:0496) and this project's scaffold protein as prey (MI:0498). Zero rows run the other way.** The six remaining rows - CAMKV and EPB41L1, already adjudicated in cycle 54 - were reported with the same direction, giving **30 of 30** [E1]. **No experiment in the entire queue was performed by anyone asking what binds KMT2A, AFF4, menin, PSIP1, AFF1 or DOT1L.** Every row is a by-product of somebody else's bait. That is the read-queue-scale version of this graph's existing finding that two thirds of the interactome is somebody else's experiment - here it is **all** of it. **Three supporting facts from the same row-by-row read** [E1]: the highest IntAct MI-score anywhere in the queue is **0.53**, below IntAct's own 0.6 high-confidence convention, and the lowest is 0.27; the hosts are HEK293T, HEK293 Flp-In and HeLa S3 with **exactly one exception**, the K562 row for YWHAG, so **one of thirty rows was obtained in a haematopoietic cell and none in a B-lineage or fusion-expressing one**; and the paired "2 publications" that admit six members to the queue are **BioPlex 2017 and BioPlex 2021 - one screening programme, two releases**. **Seven members are rejected here with measurements** [E2][E3]: FGF12, PLCD3, DENND2D, FAM9A, MDK, ARRB2 and RPS11. Two more - CAMKV and EPB41L1 - **were already rejected on 2026-09-20/21 by the cycle-54 adjudication and the read-queue instrument cannot see that**, because a rejection is not a `gene-*` node so `own_entry` stays null and they re-queue every run [E4]. Six got entries: RSBN1, PES1, AFG2A, AFG2B, YWHAG and (as the rejected substrate's proper home) `mark-h4k20me2`. **And not one of the fifteen is a selective dependency**: across the nine measured here, DepMap 26Q2 gene effects run from **+0.30 (PLCD3) to -2.84 (RPS11)** with the negatives all being pan-essentials and the leukaemia subset never materially more negative than the panel [E3].

Go deeper
The direction table - every row, read not assumed

MITAB2.7 columns 19 and 20 carry experimental role A and role B; column 29 carries the host organism; column 7 the detection method; column 15 the IntAct MI-score. All were read out per row on 2026-09-22 1:

candidatescaffoldpubmethodhostrole of candidaterole of scaffoldMI
RSBN1AFF428514442anti-tag coIPHEK293Tbaitprey0.53
RSBN1AFF433961781anti-tag coIPHEK293Tbaitprey0.53
PES1AFF428514442anti-tag coIPHEK293Tbaitprey0.53
PES1AFF433961781anti-tag coIPHEK293Tbaitprey0.53
FAM9AAFF428514442anti-tag coIPHEK293Tbaitprey0.53
FAM9AAFF433961781anti-tag coIPHEK293Tbaitprey0.53
MDKAFF428514442anti-tag coIPHEK293Tbaitprey0.53
MDKAFF433961781anti-tag coIPHEK293Tbaitprey0.53
DENND2DAFF428514442anti-tag coIPHEK293Tbaitprey0.53
DENND2DAFF433961781anti-tag coIPHEK293Tbaitprey0.53
ARRB2AFF428514442anti-tag coIPHEK293Tbaitprey0.35
ARRB2AFF433961781anti-tag coIPHEK293Tbaitprey0.35
PLCD3AFF433961781anti-tag coIPHEK293Tbaitprey0.35
PLCD3AFF133961781anti-tag coIPHEK293Tbaitprey0.35
FGF12AFF433961781anti-tag coIPHEK293Tbaitprey0.35
FGF12AFF133961781anti-tag coIPHEK293Tbaitprey0.35
YWHAGKMT2A36931259BioIDHEK293 Flp-Inbaitprey0.41
YWHAGKMT2A39251607BioIDK562baitprey0.41
AFG2AKMT2A38554706tandem affinity purificationHeLa S3baitprey0.35
AFG2AMEN138554706tandem affinity purificationHeLa S3baitprey0.35
AFG2BKMT2A38554706tandem affinity purificationHeLa S3baitprey0.35
AFG2BPSIP138554706tandem affinity purificationHeLa S3baitprey0.35
RPS11KMT2A39251607BioIDK562baitprey0.27
RPS11MEN139251607BioIDK562baitprey0.27

24 of 24. No exceptions, no partial rows, no ambiguous roles. The other six rows in the queue belong to CAMKV (4) and EPB41L1 (2), which the cycle-54 adjudication reported as bait-side in all of their rows too 4, giving 30 of 30 across the whole non-substrate queue.

What follows, and what does not. It does not follow that the rows are false - a co-purification is symmetric as physics. It follows that the read queue contains no experiment designed to enumerate this project's interactome. The PI's standing priority is to map every interactor of KMT2A::AFF1 with evidence; the residual queue after 95 entries consists entirely of rows nobody generated for that purpose, in cells that are not the disease, at confidence scores below the database's own high-confidence line.

The corollary is a specific, cheap experiment. For any queue member worth keeping, the reciprocal pulldown - tag the scaffold, ask whether the candidate comes out - has never been done, and it is the version of the experiment that would mean something. This graph already holds exp-no-proximity-labelling-has-ever-been-done-on-a-kmt2a-fusion, which is the same gap at the fusion level.

The seven rejections, with their measurements

All DepMap figures are release 26Q2 (meta.dataVersion year 26, month 06), fetched 2026-09-22, with all 29 tissue blocks concatenated; all UniProt records fetched live the same day 23.

geneaccession, aaUniProt localisation, with evidence codeDepMap mean (screens)% lines < -0.5rejection ground
FGF12P61328, 243Nucleus, ECO:0000305 ONLY - curator inference, not experiment. Function: nervous-system development, raises the voltage dependence of SCN8A fast inactivation+0.0829 (1,254)0.0%an intracellular FGF whose only function is sodium-channel modulation in neurons; the sole localisation is a curator inference; gene effect POSITIVE in every line
PLCD3Q8N3E9, 789Cell membrane, Cytoplasm, Cleavage furrow - all ECO:0000269. No nucleus of any kind+0.2989 (77 screens)0.0%a PIP2-hydrolysing phospholipase at the membrane and the cleavage furrow, essential for trophoblast and placental development. No nuclear pool, no chromatin module, positive gene effect
DENND2DQ9H6A0, 471Cytoplasm only, ECO:0000269-0.1035 (1,257)1.4%a RAB9A/RAB9B guanine-nucleotide exchange factor. Cytoplasmic vesicle trafficking; no nuclear pool, no chromatin module, no dependency
FAM9AQ8IZU1, 332Nucleus, nucleolus, ECO:0000269 (PubMed 12213195)-0.0325 (1,216)0.8%does have an experimentally coded nucleolar pool - so it is NOT rejected on localisation. Rejected because UniProt records no function whatsoever for it, it is an X-linked testis-restricted protein, and there is no dependency. Rejected on expression and absent function, and the localisation is recorded honestly against the rejection
MDKP21741, 143Secreted (uncoded)-0.1025 (1,258)1.4%midkine is a secreted cytokine/growth factor signalling through cell-surface proteoglycan receptors. A secreted protein recovered from a nuclear-complex coIP is topologically the least plausible row in the queue
ARRB2P32121, 409Cytoplasm, Nucleus (UNCODED), Cell membrane, clathrin-coated pit (ECO:0000250), cytoplasmic vesicle+0.0434 (1,258)0.2%beta-arrestin-2, a GPCR desensitisation and clathrin-adaptor protein. The nuclear comment carries no evidence code. Gene effect positive
RPS11P62280, 158Cytoplasm (ECO:0000269) and Nucleus, nucleolus (ECO:0000269, PubMed 34516797) - a genuine SSU-processome component-2.8391 (1,258)100.0%the most essential gene this beat measured. A hyper-abundant pan-essential small-subunit ribosomal protein with an experimentally coded nucleolar pool, recovered by BioID at the queue's LOWEST MI-score, 0.27. Rejected as an abundance/proximity artefact, not on localisation

Two of these rejections are deliberately not localisation arguments, and saying so matters. FAM9A and RPS11 both have experimentally coded nucleolar pools; the cheap version of this audit would have rejected them on "not nuclear" and been wrong. FAM9A is rejected because UniProt records no function for it at all and it is testis-restricted; RPS11 because a protein required by 100% of 1,258 cell lines is present in every nucleolus at high copy number and a BioID hit on it at MI-score 0.27 is what abundance looks like.

The ribosomal-protein context makes the RPS11 case, and it is a count. RPS11 is one of seven ribosomal or mitoribosomal proteins among the 122 unwritten union partners - with RPS9, RPL10, RPL13, RPL13A, RPL36 and MRPL12 - and it is in the read queue only because it reached two baits where the other six reached one each 5. The class is present; RPS11 is its threshold-crossing member, not a distinguished one.

What would reverse any of these seven: a dedicated experiment in a KMT2A-rearranged human cell in which the scaffold protein is the bait and the candidate is recovered; or a functional result - knockdown or knockout changing a fusion-driven readout - in a haematopoietic system. Neither exists for any of the seven, and the bounded literature for each is thin to empty.

Two members were rejected two days ago and the instrument re-queues them every run

CAMKV and EPB41L1 are in the current read queue with own_entry: null, and both were formally rejected with full reasoning in the cycle-54 adjudication recorded at ana-q99-read-queue-bounded-rejections 4. That entry gives CAMKV's accession, its pseudokinase status (UniProt: "Does not appear to have detectable kinase activity"), DepMap -0.0215, CAMKV[tiab] AND leukemia[tiab] = 0; and EPB41L1's Q9H4G0, its uncoded cytoskeletal localisation, DepMap +0.0087, and the finding that its two apparent leukaemia hits are a gene-name artefact.

This is an instrument defect, not a disagreement. analyses/core-complex-interactome-gap/run.py sets own_entry by matching a node id of the form gene-<symbol> or protein-<symbol>. A documented rejection is an ana-* node, so it does not match, so a rejected partner is indistinguishable from an unexamined one and re-enters the queue on every run. Two of the fifteen slots in the current queue - 13% - are already-answered questions, and a beat that took the queue at face value would redo them.

The fix is one line and it belongs in the instrument: a rejected set, read from the ana-* rejection entries or from data/symbol-collisions.json (whose block_autogeneration flag already does exactly this job for H3C1 and is already consulted by scripts/check-named-but-unwritten.py). Until then the queue's size is an overstatement by the number of prior rejections.

This beat's own position on both: the cycle-54 rejections stand. Nothing found here reverses either, and the direction audit strengthens both - CAMKV's and EPB41L1's rows are bait-side like all the others.

Not one queue member is a selective dependency

DepMap 26Q2, all 29 tissue blocks concatenated, leukaemia subset by substring match on diseaseFromSource and not stratified by KMT2A rearrangement 3:

genemeanleukaemia mean% lines < -0.5reading
RPS11-2.8391-2.8250100.0%pan-essential
AFG2B-1.3319-1.406495.5%pan-essential
PES1-1.0246-1.021496.9%pan-essential
AFG2A-0.8275-0.789990.5%near-pan-essential
YWHAG-0.2275-0.20586.0%not a dependency; 7 paralogues
DENND2D-0.1035-0.14891.4%not a dependency
MDK-0.1025-0.12101.4%not a dependency
RSBN1-0.0372-0.01920.3%not a dependency
FAM9A-0.0325-0.05110.8%not a dependency
ARRB2+0.0434+0.06720.2%not a dependency
FGF12+0.0829+0.08600.0%not a dependency
PLCD3+0.2989+0.36080.0%not a dependency (only 77 screens)

The distribution is bimodal and neither mode is useful. Four members are pan-essential housekeeping machines - a ribosomal protein, two replisome ATPases and a 60S maturation factor - where the leukaemia mean is within 0.08 of the panel mean in every case. The other eight are not dependencies at all, four of them with positive gene effects. There is no middle.

Read carefully, this is a null and it is reported as one. No proportion, no median split, no "anti-selective half" - this project has burned four cycles on that construction and its base rate is 45.3%. The claim here is absolute: the largest leukaemia-versus-panel gap in the table is 0.08 (AFG2B, in the more-dependent direction) and the largest in the other direction is 0.06 (PLCD3, on 77 screens). Nothing in this queue is selective, and the fourth consecutive beat in this project to look for a selective dependency in its own gene set has not found one.

PLCD3's screen count is a rule-83 case worth flagging separately: 77 screens against 1,254-1,258 for every other gene in the table, from 17 tissue blocks rather than 29. DepMap screens each gene in a gene-specific number of lines, so this is legitimate variation and not a mixed release - but a mean drawn from 77 lines is not comparable in precision to one drawn from 1,258, and the figure should never be quoted beside the others without its denominator.

One isoform trap, and it is a rediscovery of a live class

In the BioPlex 2017 row, ARRB2's participant identifier is uniprotkb:P32121-3 - isoform 3 - while in the BioPlex 2021 row it is the canonical uniprotkb:P32121 1. The two releases deposited the same interaction against two different identifiers for the same protein.

Consequence: a query or filter keyed on the canonical accession sees one of ARRB2's two rows. A union keyed on the resolved gene symbol, as this project's instrument uses, sees both - which is the right design and is why the queue reports intact_rows: 2.

This is a rediscovery, and rule 88's corollary says to say so. The cycle-54 adjudication found the identical failure for CAMKV, whose IntAct record is split between the canonical accession and isoform 2, and recorded it as registry-grade. Finding it again one cycle later on a different protein, in the same publication pair, establishes that it is systematic in BioPlex curation rather than incidental - which is a stronger claim than either finding alone, and it means every accession-keyed count against BioPlex rows in this project is a floor. The registry (147 rows, 2026-09-22) was checked for ARRB2 and for P32121 before this was written: no row for either.

Facts and figures
scriptnone - direct IntAct MITAB2.7, UniProt REST and Open Targets GraphQL queries, read row by row
statusrun
headline24 of 24 inspected read-queue rows are prey-side for this project's proteins (30 of 30 including the two cycle-54-adjudicated members, whose direction is carried second-hand). Zero rows come from an experiment that tagged KMT2A, AFF4, MEN1, PSIP1, AFF1 or DOT1L.
epigenetic readoutOnly one of the fifteen queue members is a chromatin enzyme of any kind: RSBN1/KDM9, an H4K20me2/me3 demethylase, written by this beat along with mark-h4k20me2. None of the other fourteen writes, reads or erases any histone modification, and none has a chromatin-reader module. The read queue's epigenetic content is therefore one enzyme out of fifteen, and the one substrate row (H3C1) is a mark question misfiled as a partner question.
cell typeMeasured across the whole queue and recorded as a finding: HEK293T 16 rows, HeLa S3 4, K562 3, HEK293 Flp-In 1. THREE of thirty rows were obtained in a haematopoietic cell, all K562 (CML-derived erythroleukaemia, not KMT2A-rearranged, not B-lineage). ZERO rows in a B-lineage cell, zero in an infant cell, zero in a primary leukaemia sample, zero in a fusion-expressing cell.
cell stateNo queue row carries a cell-state condition at all - every one is an unperturbed proliferating culture. No row is from a differentiation-blocked, drug-treated, replication-stressed or otherwise conditioned state, so the queue contains no state-resolved interaction data of any kind.
localizationMeasured per member as part of the rejection audit and recorded in data.rejections. Of the seven rejected, three have NO nuclear pool annotated (PLCD3, DENND2D, MDK-secreted), two have a nuclear pool that is not experimentally coded (FGF12 at ECO:0000305, ARRB2 uncoded), and TWO have experimentally coded nucleolar pools and are therefore NOT rejected on localisation (FAM9A, RPS11).
subnuclear localizationTwo of the fifteen queue members have an experimentally coded SUBNUCLEAR assignment and both are NUCLEOLAR: FAM9A (Nucleus/nucleolus, ECO:0000269, PubMed 12213195) and RPS11 (Nucleus/nucleolus, ECO:0000269, PubMed 34516797, as an SSU-processome component). PES1 is nucleolar by primary experiment (immunofluorescence, fractionation and sucrose gradients) though UniProt's comment is uncoded. Every other queue member has NO subnuclear assignment of any kind. Recorded as a measured distribution over the queue, not as a single address.
shown for fusionNO for every member of the read queue, without exception. Zero of thirty rows involve a KMT2A fusion protein, a fusion-expressing cell, a B-lineage cell or an infant sample; and in zero of thirty rows was one of this project's proteins the experimental bait.
rejections
MDK143.0MDK: 143RPS11158.0RPS11: 158FGF12243.0FGF12: 243FAM9A332.0FAM9A: 332ARRB2409.0ARRB2: 409DENND2D471.0DENND2D: 471PLCD3789.0PLCD3: 789
inputs
[
  "IntAct PSICQUIC MITAB2.7 for Q03164, O00255, O75475, Q9UHB7, P51825",
  "UniProt REST gene_exact for FGF12, PLCD3, DENND2D, FAM9A, MDK, ARRB2, RPS11",
  "Open Targets GraphQL target.depMapEssentiality, DepMap 26Q2",
  "analyses/core-complex-interactome-gap/results.json (generated 2026-09-22)",
  "data/graph/nodes/interactome-read-queue/ana-q99-read-queue-bounded-rejections.json"
]
direction audit
{
  "rows_inspected_first_hand": 24,
  "rows_candidate_is_bait": 24,
  "rows_scaffold_is_bait": 0,
  "rows_carried_second_hand": 6,
  "second_hand_source": "ana-q99-read-queue-bounded-rejections (CAMKV 4 rows, EPB41L1 2 rows, reported bait-side in all)",
  "substrate_rows_excluded": 16,
  "substrate_rows_note": "H3C1's 16 rows were excluded as substrate rows and their direction was not read; see ana-q104-h3c1-is-ten-genes-on-one-accession-and-the-substrate-inflates-the-denominator-by-2-3-points",
  "mi_score_range": [
    0.27,
    0.53
  ],
  "mi_score_high_confidence_convention": 0.6,
  "n_rows_reaching_high_confidence": 0,
  "hosts": {
    "HEK293T": 16,
    "HeLa S3": 4,
    "K562": 3,
    "HEK293 Flp-In": 1
  },
  "haematopoietic_rows": 3,
  "haematopoietic_note": "All three K562 rows - YWHAG once (PMID 36931259 is the HEK293 Flp-In one; 39251607 is K562) and RPS11 twice. K562 is CML-derived erythroleukaemia, NOT KMT2A-rearranged and NOT B-lineage.",
  "b_lineage_rows": 0,
  "fusion_expressing_rows": 0,
  "methods": {
    "anti tag coimmunoprecipitation (MI:0007)": 16,
    "tandem affinity purification (MI:0676)": 4,
    "proximity-dependent biotin identification (MI:1314)": 4
  }
}
independence audit
{
  "bioplex_pair": [
    "28514442 (Nature 2017)",
    "33961781 (Cell 2021)"
  ],
  "bioplex_note": "One screening programme, same group, two releases, same HEK293T host, same method. Six queue members are admitted by this pair as '2 publications'.",
  "genuinely_two_studies": [
    "YWHAG - PMID 36931259 (Segal, Mol Cell 2023, 14-3-3 interactome, HEK293 Flp-In) and PMID 39251607 (Khoroshkin, Nat Commun 2024, K562 50-RBP BioID2) - different labs, different questions, different cells"
  ],
  "single_experiment_members": [
    "AFG2A and AFG2B - both admitted by two BAITS that are two prey proteins in ONE HeLa S3 tandem affinity purification",
    "RPS11 - two baits, one BioID experiment"
  ]
}
rejections — the numbers behind the chart
[
  {
    "gene": "FGF12",
    "uniprot": "P61328",
    "length_aa": 243,
    "localization": "Nucleus, ECO:0000305 ONLY (curator inference)",
    "function": "nervous-system development; positive regulation of SCN8A sodium-channel fast inactivation",
    "depmap_mean": 0.0829,
    "depmap_screens": 1254,
    "pct_below_minus_0_5": 0,
    "rows": 2,
    "scaffolds": [
      "AFF4",
      "AFF1"
    ],
    "pubs": [
      "33961781"
    ],
    "mi_score": 0.35,
    "ground": "an intracellular FGF whose only annotated function is neuronal sodium-channel modulation; sole localisation is a curator inference; POSITIVE gene effect in every one of 1,254 lines; both rows bait-side from one BioPlex release"
  },
  {
    "gene": "PLCD3",
    "uniprot": "Q8N3E9",
    "length_aa": 789,
    "localization": "Cell membrane, Cytoplasm, Cleavage furrow (all ECO:0000269). NO nucleus.",
    "function": "hydrolyses PIP2 to DAG and IP3; essential for trophoblast and placental development; may participate in cytokinesis at the cleavage furrow",
    "depmap_mean": 0.2989,
    "depmap_screens": 77,
    "pct_below_minus_0_5": 0,
    "rows": 2,
    "scaffolds": [
      "AFF4",
      "AFF1"
    ],
    "pubs": [
      "33961781"
    ],
    "mi_score": 0.35,
    "ground": "a membrane and cleavage-furrow phospholipase with no nuclear pool and no chromatin module; positive gene effect; both rows bait-side from one BioPlex release. NOTE the 77-screen denominator - not comparable in precision to the rest of the table"
  },
  {
    "gene": "DENND2D",
    "uniprot": "Q9H6A0",
    "length_aa": 471,
    "localization": "Cytoplasm only, ECO:0000269",
    "function": "guanine-nucleotide exchange factor activating RAB9A and RAB9B",
    "depmap_mean": -0.1035,
    "depmap_screens": 1257,
    "pct_below_minus_0_5": 0.014,
    "rows": 2,
    "scaffolds": [
      "AFF4"
    ],
    "pubs": [
      "28514442",
      "33961781"
    ],
    "mi_score": 0.53,
    "ground": "a Rab GEF in vesicle trafficking; cytoplasm only; no dependency; one bait, and the two publications are one BioPlex programme"
  },
  {
    "gene": "FAM9A",
    "uniprot": "Q8IZU1",
    "length_aa": 332,
    "localization": "Nucleus, nucleolus - ECO:0000269, PubMed 12213195. NOT rejected on localisation.",
    "function": "NONE - UniProt records no function comment at all",
    "depmap_mean": -0.0325,
    "depmap_screens": 1216,
    "pct_below_minus_0_5": 0.008,
    "rows": 2,
    "scaffolds": [
      "AFF4"
    ],
    "pubs": [
      "28514442",
      "33961781"
    ],
    "mi_score": 0.53,
    "ground": "HAS an experimentally coded nucleolar pool, so the localisation test does not reject it. Rejected because UniProt records no function whatsoever, it is an X-linked testis-restricted protein, there is no dependency, and both rows are bait-side from one BioPlex programme"
  },
  {
    "gene": "MDK",
    "uniprot": "P21741",
    "length_aa": 143,
    "localization": "Secreted (uncoded)",
    "function": "secreted cytokine and growth factor signalling through cell-surface proteoglycan and non-proteoglycan receptors",
    "depmap_mean": -0.1025,
    "depmap_screens": 1258,
    "pct_below_minus_0_5": 0.014,
    "rows": 2,
    "scaffolds": [
      "AFF4"
    ],
    "pubs": [
      "28514442",
      "33961781"
    ],
    "mi_score": 0.53,
    "ground": "midkine is SECRETED. A secreted extracellular growth factor recovered from a nuclear-complex co-immunoprecipitation is the most topologically implausible row in the queue; no nuclear pool is annotated; no dependency"
  },
  {
    "gene": "ARRB2",
    "uniprot": "P32121",
    "length_aa": 409,
    "localization": "Cytoplasm, Nucleus, Cell membrane, Cytoplasmic vesicle - ALL UNCODED; clathrin-coated pit at ECO:0000250",
    "function": "GPCR desensitisation and resensitisation; clathrin-associated sorting protein recruiting GPCRs to AP-2 in coated pits",
    "depmap_mean": 0.0434,
    "depmap_screens": 1258,
    "pct_below_minus_0_5": 0.002,
    "rows": 2,
    "scaffolds": [
      "AFF4"
    ],
    "pubs": [
      "28514442",
      "33961781"
    ],
    "mi_score": 0.35,
    "ground": "beta-arrestin-2, a GPCR/clathrin adaptor; the nucleus comment carries no evidence code; POSITIVE gene effect. THE WEAKEST OF THE SEVEN REJECTIONS - beta-arrestin nuclear functions are described in the signalling literature, so this rests mainly on the positive DepMap score and the bait-side direction",
    "isoform_trap": "P32121-3 in the 2017 release, P32121 in the 2021 release - an accession-keyed filter sees one of the two rows"
  },
  {
    "gene": "RPS11",
    "uniprot": "P62280",
    "length_aa": 158,
    "localization": "Cytoplasm (ECO:0000269) and Nucleus/nucleolus (ECO:0000269, PubMed 34516797) - a genuine SSU-processome component. NOT rejected on localisation.",
    "function": "small ribosomal subunit protein uS17; component of the SSU processome in the nucleolus",
    "depmap_mean": -2.8391,
    "depmap_screens": 1258,
    "pct_below_minus_0_5": 1,
    "rows": 2,
    "scaffolds": [
      "KMT2A",
      "MEN1"
    ],
    "pubs": [
      "39251607"
    ],
    "mi_score": 0.27,
    "ground": "required by 100.0% of 1,258 lines at a mean gene effect of -2.84 - the most essential gene this beat measured. A hyper-abundant pan-essential protein present in every nucleolus, recovered by BioID at the LOWEST MI-score in the queue (0.27), from ONE publication. Rejected as an abundance/proximity artefact. It is also one of seven ribosomal proteins among the 122 unwritten partners and is the threshold-crossing member of that class, not a distinguished one"
  }
]
already rejected and requeued
{
  "genes": [
    "CAMKV",
    "EPB41L1"
  ],
  "rejected_in": "ana-q99-read-queue-bounded-rejections (cycle 54, 2026-09-20/21)",
  "why_they_requeue": "run.py sets own_entry by matching a node id of the form gene-<symbol> or protein-<symbol>; a rejection is an ana-* node, so own_entry stays null and a rejected partner is indistinguishable from an unexamined one",
  "share_of_the_queue": "2 of 15 = 13%",
  "this_beat_position": "BOTH REJECTIONS STAND. Nothing found here reverses either, and the direction audit strengthens both.",
  "suggested_fix": "the queue filter should exclude anything ADJUDICATED, read either from the ana-* rejection entries or from data/symbol-collisions.json, whose block_autogeneration flag already does this job for H3C1 and is already consulted by scripts/check-named-but-unwritten.py"
}
entries written by this beat
[
  "gene-rsbn1",
  "gene-pes1",
  "gene-afg2a",
  "gene-afg2b",
  "gene-ywhag",
  "mark-h4k20me2"
]
queue disposition
{
  "written": 5,
  "rejected_here": 7,
  "already_rejected_in_cycle_54": 2,
  "rejected_as_substrate_in_the_registry_on_2026_09_20": 1,
  "total": 15,
  "note": "5 + 7 + 2 + 1 = 15. Every member of the read queue is now either an entry or a documented rejection."
}
dependency verdict
{
  "release": "DepMap 26Q2 (meta.dataVersion year 26 month 06), fetched 2026-09-22",
  "n_genes_measured": 12,
  "range_mean_gene_effect": [
    -2.8391,
    0.2989
  ],
  "pan_essential": [
    "RPS11 -2.8391",
    "AFG2B -1.3319",
    "PES1 -1.0246",
    "AFG2A -0.8275"
  ],
  "not_a_dependency": [
    "YWHAG -0.2275",
    "DENND2D -0.1035",
    "MDK -0.1025",
    "RSBN1 -0.0372",
    "FAM9A -0.0325",
    "ARRB2 +0.0434",
    "FGF12 +0.0829",
    "PLCD3 +0.2989"
  ],
  "max_leukaemia_versus_panel_gap": 0.08,
  "verdict": "BIMODAL AND NEITHER MODE IS USEFUL. Four pan-essential housekeeping machines with leukaemia means within 0.08 of the panel, eight non-dependencies of which four have POSITIVE gene effects, and no middle. Not one member of the queue is a selective dependency.",
  "statistical_discipline": "Reported as absolute gene effects and per-gene leukaemia-versus-panel differences. NO median-split proportion and no 'anti-selective half' tally is reported, per rule 84 - half of any gene set is above the median of 17,728 by construction and this graph's base rate is 45.3%.",
  "rule_83_flag": "PLCD3 was screened in 77 lines across 17 tissue blocks against 1,216-1,258 lines across 29 blocks for the other eleven. This is legitimate gene-specific variation, not a mixed release, but the mean is not comparable in precision and must not be quoted beside the others without its denominator."
}
identifier traps
[
  "ARRB2 is deposited as P32121-3 (isoform 3) in BioPlex 2017 and as P32121 (canonical) in BioPlex 2021. An accession-keyed filter sees ONE of the two rows; a symbol-keyed union sees both, which is why this project's instrument reports 2. REDISCOVERY: the cycle-54 adjudication found the identical split for CAMKV between its canonical accession and isoform 2, in the same publication pair - so the failure is SYSTEMATIC in BioPlex curation, not incidental, and every accession-keyed count against BioPlex rows in this project is a floor. No row for ARRB2 or P32121 in data/symbol-collisions.json (147 rows, 2026-09-22).",
  "MITAB2.5 - the format this project's instrument requests - does NOT carry the experimental-role, host-organism or MI-score columns. MITAB2.7 does. That is why the direction of the entire read queue had never been read: the instrument cannot see it."
]
How this could bite us back
  • If the entire interactome queue is prey-side, then the project's interactor census is an artefact of which proteins other laboratories chose to tag. That is a sampling bias with a direction: proteins that are popular baits elsewhere are over-represented, and proteins nobody tags are invisible regardless of biology. The coverage percentage cannot detect this.
  • Seven documented rejections raise the risk of a false negative in exactly the place it is hardest to notice. ARRB2 is the flagged case - beta-arrestin nuclear functions are described in the signalling literature, and the rejection rests mainly on a positive DepMap score. If any of the seven is wrong, nothing in this graph will contradict it, because a rejection closes a line of work and no future beat will run to find the counter-example.
  • The four pan-essential members are a trap for a target-finding reading of this queue: RPS11, AFG2B, PES1 and AFG2A all look like strong hits in any enrichment analysis precisely because they are required everywhere. An interactome-derived target list that is not filtered for pan-essentiality will surface them first.
  • Recording that no row was obtained in a B-lineage or fusion-expressing cell also means the queue cannot be used to argue an absence. A partner that requires the infant B-cell proteome or the fusion itself cannot appear in HEK293T, HeLa S3 or K562, so every negative in this queue is conditional on host and none of them bounds the real interactome.
Still open
  1. Run the reciprocal pulldown for the members that were written: tag AFF4 in a KMT2A::AFF1 line and ask whether RSBN1 or PES1 comes out. No scaffold-as-bait experiment exists for any queue member, so this is the single missing measurement for the whole queue and it is one immunoprecipitation per candidate.
  2. 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.
  3. Add an adjudicated set to the read-queue instrument so that documented rejections stop re-queueing. Two of fifteen slots are currently already-answered questions, and the mechanism that makes them invisible is a regex on node ids.
  4. Should the instrument request MITAB2.7 instead of 2.5? The role, host and MI-score columns would let the queue be sorted by whether anyone has ever used a scaffold as bait, which is a far better priority order than bait count.
  5. How many of the 95 partners WITH entries rest on prey-side rows only? This beat measured the direction of the residual queue. If the same audit over the written 95 finds the same pattern, then the statement 'this project has 95 documented interactors' needs the same qualification the queue does.
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Reading MITAB2.7 experimental-role columns 19 and 20, host column 29, detection-method column 7 and MI-score column 15 for every read-queue row against the six scaffold accessions, all 24 rows covering the thirteen non-substrate queue members inspected have the candidate as experimental bait (MI:0496) and the scaffold protein as prey (MI:0498), with zero rows in the opposite direction; the MI-scores range from 0.27 (RPS11) to 0.53 and none reaches 0.6; the hosts are HEK293T for all sixteen BioPlex rows, HEK293 Flp-In for one YWHAG row, HeLa S3 for all four 55LCC rows and K562 for three rows (YWHAG once, RPS11 twice); and in PMID 28514442 ARRB2's participant identifier is uniprotkb:P32121-3 while in PMID 33961781 it is uniprotkb:P32121.

    What this evidence cannot show

    Queried live 2026-09-22 in MITAB2.7 (the project's instrument uses tab25, which does NOT carry the role and host columns - that is why this had never been read). Filtering was by ACCESSION appearing in the idA or idB field, not by alias token, so an isoform-suffixed identifier such as P32121-3 still matched. WHAT IS NOT COVERED: the six rows belonging to CAMKV (4) and EPB41L1 (2) were NOT re-inspected by this beat - their direction is carried from the cycle-54 adjudication, which reported the rejected protein as bait in all its rows, and the 30-of-30 figure therefore rests on 24 first-hand and 6 second-hand. H3C1's 16 rows were excluded as substrate rows and their direction was not read. A BAIT/PREY ROLE IS A CURATION FIELD: it records what the depositors declared, and for a proteome-scale programme it is a property of the screen design rather than of the molecules. Rule 93 endpoint caveat: PSICQUIC returns 651 of the 1,015 rows IntAct REST gives for these six accessions, so rows outside the instrument were not inspected at all.

    Evidence that argues against this (1)
    • A co-purification is physically symmetric, so 'the candidate is the bait' is a statement about experimental design and not about the molecules. It does not make any row false. The inference it supports is narrower and is the one made here: no experiment in the queue was designed to enumerate this project's interactome, so the reciprocal test has never been performed for any of them. Huttlin EL et al. 2021
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    UniProt records fetched live for the seven rejected candidates give: FGF12 (P61328, 243 aa) a sole Nucleus location at ECO:0000305 with function in nervous-system development and positive regulation of SCN8A sodium-channel fast inactivation; PLCD3 (Q8N3E9, 789 aa) cell membrane, cytoplasm and cleavage furrow all at ECO:0000269 and NO nuclear location, hydrolysing PIP2 and essential for trophoblast and placental development; DENND2D (Q9H6A0, 471 aa) cytoplasm only at ECO:0000269, a RAB9A/RAB9B guanine-nucleotide exchange factor; FAM9A (Q8IZU1, 332 aa) Nucleus/nucleolus at ECO:0000269 (PubMed 12213195) and NO function comment at all; MDK (P21741, 143 aa) Secreted, a cytokine and growth factor signalling through cell-surface proteoglycan receptors; ARRB2 (P32121, 409 aa) cytoplasm, nucleus, cell membrane and cytoplasmic vesicle all UNCODED plus clathrin-coated pit at ECO:0000250, a GPCR desensitisation and clathrin-adaptor protein; and RPS11 (P62280, 158 aa) 'Small ribosomal subunit protein uS17', cytoplasm at ECO:0000269 and Nucleus/nucleolus at ECO:0000269 (PubMed 34516797) as an SSU-processome component.

    What this evidence cannot show

    Fetched live 2026-09-22; every one of the seven symbols returned EXACTLY ONE reviewed human record, so rule 62 is satisfied for all of them. Evidence codes were read from the JSON rather than inferred, and the distinction between ECO:0000269 (experimental), ECO:0000305 (curator inference) and no code at all is preserved per rule 91 - and NONE of the cited papers was opened, so in every case the code establishes that a curator read an experiment, not that the paper is about the location. TWO OF THE SEVEN HAVE EXPERIMENTALLY CODED NUCLEOLAR POOLS - FAM9A and RPS11 - and are therefore expressly NOT rejected on localisation; rejecting them on 'not nuclear' would have been wrong and is the error this item exists to prevent. The FAM9A function absence is an absence in UniProt, not a demonstration that the protein has no function.

    Evidence that argues against this (1)
    • An uncoded or inferred localisation is an annotation gap, not a demonstration of absence. ARRB2 in particular is known in the signalling literature to act in the nucleus, and beta-arrestin nuclear functions including histone-acetyltransferase recruitment have been described - so rejecting ARRB2 partly on an uncoded nucleus comment is the weakest of the seven rejections and rests mainly on its positive DepMap gene effect and on the bait-side direction of both its rows. UniProt Knowledgebase 2026
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    In DepMap release 26Q2 as served by Open Targets, the twelve read-queue members measured have mean gene effects of RPS11 -2.8391 (1,258 screens, 100.0% of lines below -0.5), AFG2B -1.3319 (1,258, 95.5%), PES1 -1.0246 (1,258, 96.9%), AFG2A -0.8275 (1,258, 90.5%), YWHAG -0.2275 (1,258, 6.0%), DENND2D -0.1035 (1,257, 1.4%), MDK -0.1025 (1,258, 1.4%), RSBN1 -0.0372 (1,258, 0.3%), FAM9A -0.0325 (1,216, 0.8%), ARRB2 +0.0434 (1,258, 0.2%), FGF12 +0.0829 (1,254, 0.0%) and PLCD3 +0.2989 (77 screens, 0.0%); and in every case the leukaemia-annotated subset mean differs from the whole-panel mean by at most 0.08.

    What this evidence cannot show

    RELEASE RECORDED per rule 83: the same API call returned meta.dataVersion = {year: 26, month: 06} = DepMap 26Q2, fetched 2026-09-22. SCREEN COUNTS ARE GENE-SPECIFIC and differ legitimately - 1,216 to 1,258 for eleven genes and 77 for PLCD3, which came from 17 tissue blocks rather than 29. THE PLCD3 MEAN IS NOT COMPARABLE IN PRECISION to the others and must never be quoted beside them without its denominator. All tissue blocks were concatenated before averaging; reading block [0] returns about 2% of screens and is a registered trap. Every Ensembl id was resolved through the HGNC primary ensembl_gene_id per rule 78 and every readback control passed - Open Targets returned the expected approvedSymbol for all twelve, including the migrated AFG2A and AFG2B. THE LEUKAEMIA SUBSET IS NOT A KMT2A SUBSET: it is a substring match on diseaseFromSource for 'leukemia', 'leukaemia', 'lymphoblastic' or 'myeloid', 54-57 screens, and NO KMT2A-rearrangement stratification was performed - so nothing here can speak to fusion selectivity in either direction. NO MEDIAN-SPLIT OR 'ANTI-SELECTIVE HALF' PROPORTION IS REPORTED, per rule 84: half of any gene set is above the median of 17,728 by construction and this graph's base rate is 45.3%.

    Evidence that argues against this (1)
    • A CRISPR viability screen is the wrong instrument for most of this queue and the table should not be read as evidence that these proteins are unimportant. YWHAG has six paralogues, RSBN1 has RSBN1L, and a flat single-gene score in a redundant family is uninformative in both directions. For the pan-essentials the opposite problem applies: a gene required by 90-100% of lines cannot show selectivity even if a fusion-specific role exists, because there is no viable comparison group. Open Targets Platform / DepMap 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    CAMKV and EPB41L1 appear in the current read queue with own_entry null although both were formally rejected with reasoning, accessions, DepMap figures and bounded literature counts in this graph's cycle-54 adjudication entry ana-q99-read-queue-bounded-rejections; the cause is that analyses/core-complex-interactome-gap/run.py sets own_entry by matching a node id of the form gene-<symbol> or protein-<symbol>, so a rejection recorded as an ana-* node cannot be seen and a rejected partner is indistinguishable from an unexamined one on every subsequent run.

    What this evidence cannot show

    Both files read directly by this beat on 2026-09-22. The own_entry mechanism was read in run.py's load_graph() - it builds own_entry from a regex on node ids and the queue filter is 'not r["own_entry"]', which is the whole of the defect. THE URL IS A PLACEHOLDER: both sources are files in this repository with no external URL; the paths are the identifiers and are given in the title. THE CYCLE-54 REJECTION FIGURES ARE QUOTED FROM THAT ENTRY AND WERE NOT RE-DERIVED by this beat - CAMKV's DepMap -0.0215 and EPB41L1's +0.0087 are that entry's numbers on 1,208 lines, which is a DIFFERENT screen count from the 1,258 this beat sees in 26Q2, so they are not directly comparable to the table in E3 and were deliberately not merged into it.

    Evidence that argues against this (1)
    • The instrument's behaviour is defensible as designed: own_entry means 'has an entry of its own', which is what the PI's one-entry-per-thing instruction asks about, and a rejection genuinely is not an entry. The defect is therefore in the QUEUE FILTER rather than in the field - the queue should exclude anything adjudicated, whether written or rejected, and the field should keep meaning what it says. this repository 2026
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    RPS11 is one of seven ribosomal or mitochondrial-ribosomal proteins among the 122 union partners with no entry - the others being RPS9, RPL10, RPL13, RPL13A, RPL36 and MRPL12 - and it reaches the read queue only because it appears against two baits (KMT2A and MEN1) where each of the other six appears against one; no ribosomal-protein gene entry of any kind exists in this graph.

    What this evidence cannot show

    Computed by this beat 2026-09-22 over the 122-member all_missing list. THE CLASS WAS MATCHED BY REGEX on ^(RPS|RPL|MRPS|MRPL)[0-9], so any ribosomal-subunit protein under a non-conforming approved symbol is uncounted - RACK1, a 40S component, is the obvious example - making seven a FLOOR for the class. The absence of ribosomal gene entries was checked by regex over the complete 2,019-id node listing. The per-gene bait counts come from results.json's own fields and were not re-derived from raw MITAB. The union comes from PSICQUIC, which returns 64% of the rows IntAct REST gives for these accessions.

    Evidence that argues against this (1)
    • Ribosomal proteins are not automatically artefacts in this context and the counter-case is real: this graph holds an eleven-member nucleolar block from a dedicated MLL-fusion AP-MS study, the fusion is reported as predominantly nucleolar, and RPS11 has an experimentally coded nucleolar SSU-processome role. The rejection therefore rests on abundance and on the MI-score of 0.27 - the lowest in the queue - and not on ribosomal proteins being intrinsically uninteresting here. UniProt Knowledgebase 2026
Connected to