Reading the fusion interactome by COMPLEX rather than by protein — and the strongest signature in it is a known contaminant, which is what makes the rest readable

{
  "pmid": "29777171",
  "network": "960 proteins, Supplementary Table 5, seven MLL-fusion AP-MS baits in isogenic Jurkat"
}

[
  {
    "complex": "CCT/TRiC",
    "found": "8/9",
    "p": 2.4e-10,
    "pan_fusion": [],
    "role": "NEGATIVE CONTROL — canonical AP-MS contaminant"
  },
  {
    "complex": "SEC",
    "found": "10/16",
    "p": 3.8e-10,
    "pan_fusion": [],
    "role": "POSITIVE CONTROL — present and correctly bait-specific"
  },
  {
    "complex": "HBO1/KAT7",
    "found": "6/10",
    "p": 0.0000021,
    "pan_fusion": [
      "KAT7",
      "JADE3"
    ]
  },
  {
    "complex": "BAF/SWI-SNF",
    "found": "6/17",
    "p": 0.000094,
    "pan_fusion": [
      "SMARCC1"
    ]
  },
  {
    "complex": "CPSF/3'-end",
    "found": "5/12",
    "p": 0.00015,
    "pan_fusion": [
      "NUDT21",
      "WDR33"
    ]
  },
  {
    "complex": "U2 snRNP",
    "found": "5/12",
    "p": 0.00015,
    "pan_fusion": [
      "SF3A3"
    ]
  },
  {
    "complex": "NuRD",
    "found": "5/14",
    "p": 0.00035,
    "pan_fusion": [
      "CHD4"
    ]
  },
  {
    "complex": "Nucleoporins",
    "found": "5/27",
    "p": 0.0084,
    "pan_fusion": [
      "TPR",
      "NUP50"
    ]
  },
  {
    "complex": "MCM2-7",
    "found": "2/6",
    "p": 0.03,
    "pan_fusion": [
      "MCM2"
    ]
  },
  {
    "complex": "Elongin",
    "found": "1/6",
    "p": 0.26,
    "pan_fusion": [
      "TCEB3"
    ],
    "role": "not significant"
  }
]

{
  "was_TPR_nuclear_envelope_carry_over": "NOT SUPPORTED by a complex-wide signature. The nucleoporins are the weakest chromatin-adjacent row here — 5 of 27, p = 8.4e-3, against 1.30 expected. A carry-over explanation predicts the pore complex arriving broadly; it did not. This does not make TPR a specific partner, but the cheapest explanation for it is not available.",
  "is_JADE3_a_real_module_member": "YES, on this evidence. HBO1/KAT7 is the best-supported non-control complex at 6 of 10 subunits with BOTH KAT7 and JADE3 pan-fusion, and it converges with the one independent non-Skucha contact in the beat — PMID 34431785, MLL fusions contacting HBO1 through THD2 'mediated mainly by the ING4/5 and PHF16 subunits', where PHF16 is JADE3's synonym and ING4 is also in the 960. CONFIRMED per-bait: HBO1/KAT7 is 4 of 10 in the MLL-AF4 column specifically, with BRPF1, JADE2, JADE3 and KAT7 present and 3 of the 4 at p<0.05. This is the only chromatin complex besides the SEC that survives the per-bait test for the index fusion.",
  "were_NuRD_and_BAF_lone_subunits": "IT DEPENDS ON THE DENOMINATOR, and I answered this wrongly before running the per-bait pass. ACROSS THE POOLED NETWORK: no -- NuRD brings 5 of 14 and BAF 6 of 17, so the argument from loneliness fails at network level. IN THE MLL-AF4 COLUMN, which is the one that matters for this project: NuRD is CHD4 ALONE (1 of 14) and BAF is 3 of 17 with only one subunit at p<0.05. So gene-smarcc1's original caution was right about the index fusion and I over-withdrew it. The pooled figures come from other baits -- MLL-AF9 and MLL-EEN supply 4 NuRD subunits each, MLL-CBP supplies 6 BAF subunits. See data.per_bait_added_same_day.",
  "is_MCM2_a_replication_licensing_story": "NO. Only MCM2 and MCM3 of six, p = 0.03. The MCM2-7 hexamer did not arrive.",
  "is_TCEB3_an_Elongin_story": "NO. It is the ONLY Elongin subunit present, 1 of 6, p = 0.26, not significant. Elongin A in an elongation-driven fusion is a tempting reading with no complex support behind it."
}

{
  "why": "The ranking above pools all seven baits, and a union is not a statement about any one of them. This was open question 4 in this entry; answering it changed a conclusion.",
  "index_fusion_column_MLL_AF4": {
    "SEC": "10/16, 9 at p<0.05 -- the dominant real module for this fusion",
    "CCT/TRiC": "8/9, 6 significant -- the contaminant is fully present per-bait too",
    "HBO1/KAT7": "4/10, 3 significant (BRPF1, JADE2, JADE3, KAT7) -- HOLDS UP",
    "CPSF/3'-end": "4/12, 2 significant (CPSF4, CSTF2, NUDT21, WDR33)",
    "BAF/SWI-SNF": "3/17, only 1 significant",
    "U2 snRNP": "2/12, 1 significant",
    "Nucleoporins": "2/27 (NUP50, TPR), both significant",
    "NuRD": "1/14 -- CHD4 ALONE",
    "MCM2-7": "2/6",
    "Elongin": "1/6"
  },
  "the_per_bait_positive_control_is_stronger_than_the_pooled_one": "SEC is 10/16 in MLL-AF4 and 10/16 in MLL-AF9, but 0/16 in MLL-AF1p and MLL-EEN, 1/16 in MLL-CBP and MLL-GAS7, and 2/16 in MLL-ENL. Those two baits are precisely the fusions whose C-terminal partners (AFF1/AF4 and MLLT3/AF9) are SEC components. The method recovers that pattern without being told it, which is a far better validation than the pooled SEC row, where the enrichment was close to tautological.",
  "what_this_REVERSES": "I used the pooled NuRD and BAF figures earlier today to withdraw the 'lone core subunit' caution in gene-smarcc1 and gene-chd4. Per bait that withdrawal is wrong for the index fusion: NuRD in the MLL-AF4 column is CHD4 alone, and BAF is three subunits with one significant. The withdrawal was right about the 960-protein network and wrong about the column that matters. Both entries and the edge now carry the correction.",
  "what_this_STRENGTHENS": "HBO1/KAT7 survives the stricter per-bait test in the index fusion's own column with 3 of its 4 subunits significant, which together with the retained THD2 domain and the independent Takahashi contact makes it the best-evidenced module in this beat."
}

{
  "why_tested": "Q87 found AQR in ECAS (PMID 42748909, Mol Cell 2026) -- 'a bipartite assembly linking an elongation module (AFF4-ELL2-EAF1) to a U2 snRNP-associated intron binding complex ... molecularly distinct from canonical P-TEFb-containing SEC'. I speculated in a commit message that the U2 snRNP signal in this network might be ECAS rather than spliceosome background, and therefore a second elongation assembly the fusion could engage.",
  "result": "REFUTED by the data I already had.",
  "ECAS elongation module (AFF4-ELL2-EAF1)": "1 of 3 in the 960. ELL2 and EAF1 are ABSENT from the network entirely. The one present, AFF4, is a fusion partner and is in the network for that reason.",
  "intron binding complex (AQR, XAB2, ISY1, ZNF830, PPIE, BUD13, RBM22, CDC40)": "2 of 8 in the 960 (BUD13, XAB2), and NEITHER is in the MLL-AF4 column. AQR itself -- the protein that prompted the question -- is ABSENT from the 960.",
  "canonical SEC core for comparison": "7 of 8 in the 960, all 7 in the MLL-AF4 column.",
  "conclusion": "The U2 snRNP subunits in this interactome are not ECAS. ECAS's defining components are absent. My speculation was wrong and is withdrawn; the fusion engages the canonical SEC, and the spliceosome signal remains unexplained rather than reassigned.",
  "what_this_does_NOT_touch": "Q87's AQR finding stands on its own evidence -- AQR's presence in ECAS on promoter-proximal chromatin is a claim about chromatin occupancy from a different experiment, not a claim that AQR binds the fusion. The two are simply not connected by this dataset, which is worth stating because they look connected."
}

{
  "what_changed": "Added two complexes Q86 flagged as untested -- NEXT/exosome and Nucleolar/ribosome biogenesis -- and Benjamini-Hochberg across all twelve, which cycle 48 owed and which the entry admitted the weakest rows would not survive.",
  "the_nucleolar_row_is_now_the_top_of_the_table": "15 of 26 subunits, p = 7e-14, q(BH) = 8.4e-13, against 1.25 expected. That is ahead of BOTH controls -- CCT/TRiC q = 1.4e-9 and SEC q = 1.5e-9. Members found: FBL, FTSJ3, GNL2, LYAR, NOLC1, NOP2, NOP58, PA2G4, RRP12, RRP9, TBL3, TCOF1, UTP14A, UTP18, WDR18.",
  "AND_THE_INDEX_FUSION_HAS_THE_FEWEST": "Per-bait subunit counts out of 26: MLL-AF1p 8, MLL-AF4 7, MLL-AF9 12, MLL-CBP 10, MLL-EEN 12, MLL-ENL 11, MLL-GAS7 7. MLL-AF4 is tied last, with 7 subunits in its column and only 4 at p<0.05 (GNL2, NOP2, PA2G4, TBL3). The strongest complex-level signal in the whole interactome is WEAKEST in the fusion this project studies.",
  "why_that_matters_for_the_nucleolar_convergence": "The standing convergence rests partly on MLL-AF4 being reported predominantly nucleolar. If nucleolar residence were driving the co-purification, the index fusion should pull MORE nucleolar protein than the others. It pulls the least. The competing explanation is abundance -- the nucleolus is the densest compartment in the nucleus and every bait is a tagged nuclear protein -- and a top-ranked hypergeometric hit against a whole-proteome background is exactly what abundance produces. That is the CCT/TRiC lesson applying to the biggest row in the table. Against it: all thirteen of the originally-named nucleolar proteins have a CLEAN negative control (0).",
  "NEXT_exosome": "9 of 17 pooled, q = 6.7e-8, but only 2 of 17 in the MLL-AF4 column (EXOSC4, ZC3H18). Another pooled-strong, per-bait-weak row.",
  "benjamini_hochberg": "Applied across the twelve complexes. Nothing changed rank; the Elongin row is now explicitly marked ns (q = 0.26) and MCM2-7 survives only marginally (q = 0.033)."
}

{
  "_added": "king, cycle 49. This entry flagged the background as its known weakness twice and deferred fixing it twice. Fixed.",
  "the_problem": "A nuclear-lysate AP-MS cannot sample the whole proteome, so a ~20,000 background overstates enrichment for every complex of abundant nuclear proteins. The honest denominator is the ~4,600 proteins Skucha recovered before filtering, and that list is unpublished.",
  "the_fix": "Bracket it. UniProt, fetched live 2026-09-21: reviewed human proteome 20,431; reviewed human NUCLEUS (SL-0191) 5,754; reviewed human NUCLEOLUS (SL-0188) 470. The nuclear count is close to the 4,600 recovered, which is what makes it a usable proxy rather than a guess. Benjamini-Hochberg applied within each column.",
  "q_values_by_background": {
    "Nucleolar/ribosome biogenesis": "6.2e-13 | 2.7e-05 | 0.00025",
    "CCT/TRiC (negative control)": "1.2e-09 | 2.7e-05 | 0.00025",
    "SEC (positive control)": "1.2e-09 | 0.00019 | 0.0014",
    "NEXT/exosome": "5.6e-08 | 0.002 | 0.01",
    "HBO1/KAT7": "4.5e-06 | 0.0058 | 0.019",
    "BAF/SWI-SNF": "0.00017 | 0.076 ns | 0.19 ns",
    "CPSF/3'-end": "0.0002 | 0.062 ns | 0.14 ns",
    "U2 snRNP": "0.0002 | 0.062 ns | 0.14 ns",
    "NuRD": "0.00042 | 0.092 ns | 0.2 ns",
    "Nucleoporins": "0.0093 | 0.52 ns | 0.75 ns",
    "MCM2-7": "0.032 | 0.32 ns | 0.44 ns",
    "Elongin": "0.25 ns | 0.67 ns | 0.75 ns",
    "_columns": "proteome 20,431 | nuclear 5,754 | recovered 4,600"
  },
  "WHAT_THIS_REMOVES": "SEVEN of twelve complexes stop being significant at a realistic background: BAF, CPSF, U2 snRNP, NuRD, nucleoporins, MCM2-7 and Elongin. The proteome denominator was manufacturing them. Only five survive, and TWO OF THOSE ARE THE CONTROLS -- so the real list of surviving non-control complexes is three: Nucleolar, NEXT/exosome and HBO1/KAT7.",
  "WHAT_THIS_STRENGTHENS": "HBO1/KAT7 survives every background (q = 0.0058 nuclear, 0.019 recovered), and it was already the best-evidenced module by the per-bait test and by THD2 retention. Three independent tests now agree on it.",
  "what_it_does_to_the_NuRD_argument": "It settles it in Q84's favour completely. Q84 cautioned that CHD4 and SMARCC1 were each effectively lone core subunits; I withdrew that on the pooled figures, then partly reinstated it on the per-bait split. At a realistic background NuRD (q = 0.092) and BAF (q = 0.076) are not significant at all. The original caution was right and I was wrong twice before arriving back at it.",
  "and_the_nucleolar_row": "Still the top row and still significant, but 2.7e-05 rather than 6.2e-13 -- eight orders of magnitude weaker, and now TIED WITH THE CONTAMINANT. Given that the index fusion has the FEWEST nucleolar subunits of the seven baits, the abundance reading is now the better-supported one."
}

{
  "_added": "king, cycle 50. analyses/nucleolar-block-depth-versus-pathway/run.py",
  "what_i_claimed_in_cycle_49": "That the nucleolar convergence fails a falsifiable prediction: if the index fusion's own nucleolar residency drove co-purification, MLL-AF4 should recover MORE of the block, and it recovers the fewest (7 of 26, tied with MLL-GAS7). I called that 'the largest available margin' and the agent endorsed and strengthened it.",
  "why_that_was_confounded": "MLL-AF4 is also the SHALLOWEST of the seven baits by total spectral counts -- 2,977 against MLL-ENL's 11,293. I had been carrying 'MLL-AF4 is the shallowest bait, so a low count is a bait property' as a caveat on individual proteins since cycle 48, and then failed to apply it to the bait-level comparison I built an argument on.",
  "the_clean_test": "Every bait lists ~293 proteins because of the rank cutoff, so depth cannot change HOW MANY it lists, only WHICH. Fraction-nucleolar is therefore a composition question. r(depth, % nucleolar) = 0.618, permutation p = 0.15 at n=7 -- underpowered, so not evidence of no effect.",
  "the_answer": {
    "MLL-AF4 observed": "2.41% of its 291 calls are nucleolar",
    "depth-predicted": "2.55%",
    "residual": "-0.14 percentage points, fourth of seven, essentially zero",
    "the_real_low_outlier": "MLL-GAS7 at -0.91 pp, on average depth"
  },
  "WITHDRAWN": "The 'index fusion recovers the fewest' argument. It recovers the fewest partly because it is the shallowest bait, and its nucleolar composition is unremarkable. This leg is uninformative about the compartment story in either direction.",
  "WHAT_STILL_STANDS": "Two legs, both unaffected by depth because both are WITHIN-bait comparisons or absolute absences. (1) The CCT/TRiC contaminant beats the block inside MLL-AF4's own column, 8/9 with 6 significant against 7/26 with 4. (2) No coherent particle is ever captured: FBL and NOP58 appear under different baits than RRP9, and WDR18's Rix1 partners PELP1 and TEX10 are absent entirely. The background sensitivity also stands -- at a realistic nuclear background the nucleolar row is exactly tied with the contaminant.",
  "so_the_conclusion_survives_on_two_of_three_legs": "and the leg that failed is the one I contributed. The agent's two observations were the load-bearing ones."
}