The histone/ribosomal shape in the read queue is three classes, not one - H2BC21 is distinguishable AND essential at -0.70, so neither existing rejection ground reaches it, and HMGA1 is a harder dependency than MEN1 or KMT2A
also called read queue class adjudication · H2BC21 is not H3C1 · HMGA1 harder than MEN1 · three classes not one · accession ambiguity versus pan-essentiality
Three more histone-like proteins turned up in the list of things we have not written about, and it looked like they could all be dismissed the same way the last one was. Measuring them showed the easy dismissal only works for one of them.
The same thing, for a scientist
Cycle 58's migration of the interactome instrument to IntAct REST put eight new partners in the read queue, and three of them - **H2BC21, H2BC10, RPL13** - looked like the shape cycle 57 already rejected twice. The temptation was to wave all three through on "same class as H3C1/RPS11". That would have been **wrong on two of the three.** Cycle 57 gave two *different* grounds and they were never separated: **H3C1** was rejected because UniProt P68431 carries **ten gene names**, so `gene-h3c1` would assert one of ten indistinguishable records; **RPS11** was rejected as the threshold-crossing member of a pan-essential class. Measured per symbol: | symbol | accession | gene names on it | lines | mean Chronos | frac < −0.5 | |---|---|---|---|---|---| | H3C1 | P68431 | **10** | 1,254 | −0.293 | 7.6% | | H2BC10 | P62807 | **5** | 1,244 | −0.133 | 1.7% | | **H2BC21** | Q16778 | **1** | 1,140 | **−0.703** | **76.1%** | | RPL13 | P26373 | 1 | 1,258 | −2.270 | **100%** | | RPS11 | P62280 | 1 | 1,258 | −2.839 | 100% | **So the accession-ambiguity ground applies to H3C1 and H2BC10 and to nothing else in the queue.** H2BC21 has its own accession as *"Histone H2B type 2-E"*, and both ribosomal proteins always did - which means **RPS11's rejection was never about ambiguity at all**, and the two grounds were being conflated. **Three answers, not one.** (1) *Accession ambiguity*: H2BC10 joins H3C1 and is adjudicated. (2) *Pan-essential ribosomal*: RPL13 at −2.27 with 100% of lines below −0.5 sits with RPS11 (−2.84) and the anchors RPL13A (−2.07), RPS6 (−2.44) and POLR2L (−2.92); adjudicated on RPS11's ground. (3) **H2BC21 is neither** - distinguishable, and a genuine dependency at −0.703 with 76% of 1,140 lines below −0.5. Neither existing ground reaches it, and it stays open. **The unplanned finding is HMGA1.** It was the one candidate in the queue I expected to be worth an entry - a high-mobility-group chromatin architectural factor with a real leukaemia literature - and it comes back at **−0.838 with 90.9% of lines below −0.5**, a *harder* dependency than **MEN1 (−0.362, 28.9%)** or **KMT2A (−0.206, 11.2%)** or **DOT1L (−0.147, 12.1%)**. The complex's own core members are milder dependencies than this candidate partner. Not one of the eight queue members is a *selective* dependency: the essential ones are broadly essential and the rest are flat - NUCKS1 −0.009, ASB7 −0.046, PIH1D1 −0.075, and CD226 **positive** at +0.169.
| script | analyses/read-queue-histone-ribosomal-class-adjudication/run.py |
|---|---|
| results | analyses/read-queue-histone-ribosomal-class-adjudication/results.json |
| status | run |
| run date | 2026-09-22 |
| not one is a selective dependency | The essential members are broadly essential and the rest are flat. No member of this queue shows a KMT2A-rearranged-versus-panel selectivity signal, and none was tested for one here - this run reports pan-panel means only. |
| depmap release note | RULE 83: the Open Targets depMapEssentiality payload carries no release identifier, so no version is implied. Absolute Chronos values move between releases; the contrasts within this single pull are the durable part. |
| absence class | measured-not-analysed |
panel declared before reading
{
"under_adjudication": [
"H2BC21",
"H2BC10",
"RPL13",
"NUCKS1",
"HMGA1",
"PIH1D1",
"ASB7",
"CD226"
],
"already_adjudicated_controls": [
"H3C1",
"RPS11"
],
"pan_essential_anchors": [
"RPL13A",
"RPS6",
"POLR2L"
],
"complex_anchors": [
"MEN1",
"KMT2A",
"DOT1L"
],
"why": "The panel is in the script as a literal and was printed before any value was read, so it cannot have been chosen to fit a result."
}the two grounds were being conflated
{
"ground_A_accession_ambiguity": "one UniProt accession carrying several gene names, so a gene-<symbol> entry would assert one of N indistinguishable records. Cycle 57 used this for H3C1 (10 names on P68431).",
"ground_B_pan_essentiality_as_class_membership": "the gene is the threshold-crossing member of a broadly essential class rather than a distinguished partner. Cycle 57 used this for RPS11.",
"what_this_run_establishes": "They are independent, and RPS11 satisfies only B - its accession P62280 carries ONE gene name. Treating 'histone/ribosomal' as a single class with a single ground is what would have mis-adjudicated H2BC21."
}measured per symbol — the numbers behind the chart
{
"H3C1": {
"accession": "P68431",
"gene_names_on_accession": 10,
"distinguishable": false,
"n_lines": 1254,
"mean": -0.2934,
"frac_below_minus_0_5": 0.0758
},
"H2BC10": {
"accession": "P62807",
"gene_names_on_accession": 5,
"gene_names": [
"H2BC4",
"H2BC6",
"H2BC7",
"H2BC8",
"H2BC10"
],
"distinguishable": false,
"n_lines": 1244,
"mean": -0.1329,
"frac_below_minus_0_5": 0.0169
},
"H2BC21": {
"accession": "Q16778",
"gene_names_on_accession": 1,
"protein_name": "Histone H2B type 2-E",
"distinguishable": true,
"n_lines": 1140,
"mean": -0.7027,
"frac_below_minus_0_5": 0.7614
},
"RPL13": {
"accession": "P26373",
"gene_names_on_accession": 1,
"distinguishable": true,
"n_lines": 1258,
"mean": -2.2696,
"frac_below_minus_0_5": 1
},
"RPS11": {
"accession": "P62280",
"gene_names_on_accession": 1,
"distinguishable": true,
"n_lines": 1258,
"mean": -2.8391,
"frac_below_minus_0_5": 1
},
"NUCKS1": {
"accession": "Q9H1E3",
"distinguishable": true,
"n_lines": 1258,
"mean": -0.0091,
"frac_below_minus_0_5": 0.0135
},
"HMGA1": {
"accession": "P17096",
"distinguishable": true,
"n_lines": 1258,
"mean": -0.8382,
"frac_below_minus_0_5": 0.9094
},
"PIH1D1": {
"accession": "Q9NWS0",
"distinguishable": true,
"n_lines": 1258,
"mean": -0.0748,
"frac_below_minus_0_5": 0.0135
},
"ASB7": {
"accession": "Q9H672",
"distinguishable": true,
"n_lines": 1258,
"mean": -0.0457,
"frac_below_minus_0_5": 0.0048
},
"CD226": {
"accession": "Q15762",
"distinguishable": true,
"n_lines": 1258,
"mean": 0.1691,
"frac_below_minus_0_5": 0
}
}controls behaved
{
"pan_essential_anchors": {
"RPL13A": "-2.0745, 100%, 394 lines (this gene genuinely has ~394 screens, which is why the payload guard is keyed to the panel MAXIMUM and not to a per-row floor)",
"RPS6": "-2.4369, 100%",
"POLR2L": "-2.9246, 100%"
},
"complex_anchors": {
"MEN1": "-0.3616, 28.9%",
"KMT2A": "-0.2058, 11.2%",
"DOT1L": "-0.1474, 12.1%"
},
"payload_guard": "panel maximum 1,258 screens, asserted > 1,000 before any row was believed"
}adjudicated by this run
{
"H2BC10": "ground A - five gene names on P62807 (H2BC4/6/7/8/10); a gene-h2bc10 entry would assert one of five indistinguishable protein records",
"RPL13": "ground B - pan-essential at -2.270 with 100% of 1,258 lines below -0.5, indistinguishable from RPL13A, RPS6, POLR2L and from the already-rejected RPS11"
}explicitly NOT adjudicated
{
"H2BC21": "NEITHER ground reaches it. Distinguishable (own accession, 'Histone H2B type 2-E') and a real dependency (-0.703, 76.1%). Stays in the queue as an open case.",
"HMGA1": "Not adjudicable on these grounds either - distinguishable, and a HARDER dependency than any core member of the complex. Stays open.",
"NUCKS1, PIH1D1, ASB7, CD226": "Flat or positive gene effect, but a non-dependency is NOT a rejection ground on its own. Cycle 57's rejections combined localisation AND UniProt function AND dependency; this run measured only accession structure and dependency, so these four remain open pending the localisation and function read.",
"why_this_matters": "Recording six of eight as still open is the honest outcome. The temptation in an adjudication pass is to clear the queue; clearing it on a ground that does not apply is how a partner gets silently dropped."
}- Nothing here nominates a target and two of the figures argue actively against doing so. RPL13 and RPS11 are pan-essential ribosomal proteins and HMGA1 is required in 91% of screened lines, so any intervention against them would be expected to damage every proliferating tissue in a child.
- The comparison 'HMGA1 is a harder dependency than MEN1' must not be used as an argument for targeting HMGA1 or against targeting menin. It compares a broad dependency with a selective one using a statistic that cannot tell them apart, and the entry says so in its own counter-evidence.
- Should an interactor encoded by several genes be written as a PROTEIN entry rather than declined? `protein-h2b-type-1-c-e-f-g-i` would satisfy the PI's one-entry-per-thing rule at the level the interaction data actually reports, and would stop the ambiguity ground from silently shrinking the coverage denominator. Neither cycle 57 nor this run proposed it.
- Is H2BC21's dependency selective? A pan-panel mean cannot distinguish a broadly required gene from one concentrated in KMT2A-rearranged lines, and the index-fusion split was not run.
- Why is H2BC21 screened in 1,140 lines when most of the panel has 1,258? Unexplained, and it is the kind of denominator difference this project has been bitten by before.
- Do NUCKS1, PIH1D1, ASB7 and CD226 survive a localisation and UniProt-function read? A flat gene effect is not a rejection ground on its own, and cycle 57's rejections combined three fields. Four of eight queue members are still open for that reason.
- Is the two-publication-or-two-bait read-queue threshold selecting for biology at all? CD226 - positive gene effect, one bait, anti-tag co-IP - cleared it.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
The accession-ambiguity rejection ground applies to two symbols in this queue and to no others, and it is independent of the pan-essentiality ground. Measured by exact-gene reviewed UniProt query, one record asserted per symbol: H3C1 sits on P68431 with TEN gene names (H3C1/2/3/4/6/7/8/10/11/12); H2BC10 sits on P62807 with FIVE (H2BC4/6/7/8/10); H2BC21 sits on its own accession Q16778 as 'Histone H2B type 2-E' with ONE; RPL13 (P26373) and RPS11 (P62280) each carry ONE. So RPS11 - rejected in cycle 57 - never satisfied the ambiguity ground at all, and the two grounds cycle 57 used were being treated as one class.
What this evidence cannot show
Gene-name multiplicity is a property of UniProt's record structure rather than of the biology - the H2B type 1-C/E/F/G/I protein really is one protein sequence encoded by five genes, so 'indistinguishable' means indistinguishable AT THE PROTEIN LEVEL. That happens to be exactly the level an affinity-purification record reports, which is why the ground holds, but it would not hold for a transcript-level or a locus-level question. A count within one database pull.
Evidence that argues against this (1)
- The ambiguity ground may be too permissive as a rejection. If the H2B type 1-C/E/F/G/I protein is a genuine partner, then declining to write it because five genes encode it leaves a real interactor unrecorded and the coverage denominator silently wrong. The defensible response is an entry keyed to the PROTEIN (protein-h2b-type-1-c-e-f-g-i) rather than to any one gene symbol, which this graph's id convention permits and which neither cycle 57 nor this run proposed.
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
H2BC21 is reached by neither existing ground: it is a distinguishable single-gene accession AND a substantial dependency, with mean Chronos gene effect -0.7027 and 76.1% of 1,140 screened lines below -0.5. That is markedly more essential than the two ambiguous histone records in the same pull (H3C1 -0.2934 at 7.6%; H2BC10 -0.1329 at 1.7%) and more essential than every core member of the complex under study (MEN1 -0.3616 at 28.9%; KMT2A -0.2058 at 11.2%; DOT1L -0.1474 at 12.1%). RPL13, by contrast, is squarely pan-essential at -2.2696 with 100% of 1,258 lines below -0.5, matching RPS11 (-2.8391), RPL13A (-2.0745), RPS6 (-2.4369) and POLR2L (-2.9246), and is adjudicated on that ground.
What this evidence cannot show
No DepMap release identifier is available from this payload (rule 83), so these absolute values are not comparable with figures quoted in other cycles and only the within-pull contrasts are durable. Chronos gene effect for a member of a redundant multigene histone family is unreliable in a known direction: a guide targeting one H2B gene leaves the paralogues intact, so a near-zero effect is expected whether or not the protein matters. That asymmetry was written into the script's caveats BEFORE the run and it cuts in H2BC21's favour - a LOW score under paralogue buffering is more informative than a high one, which makes -0.703 more notable rather than less. H2BC21 has 1,140 screens against 1,258 for most of the panel, and why it is screened in fewer lines was not established.
Evidence that argues against this (2)
- A gene effect of -0.70 with 76% of lines below threshold is an argument for dropping H2BC21 rather than writing it. On cycle 57's own reasoning that is class membership in the broadly-required set, not a distinguished partner - the difference from RPL13 is one of degree, and no selectivity test was run here to show it behaves differently in KMT2A-rearranged lines. Calling it 'open' rather than rejected may simply be deferring the same conclusion.
- The comparison to MEN1, KMT2A and DOT1L is doing rhetorical work it cannot support. Those three are mild pan-panel dependencies precisely because their dependency is SELECTIVE - concentrated in KMT2A-rearranged lines and diluted across 1,258 mostly-irrelevant ones. A pan-panel mean is the wrong statistic for comparing a selective dependency with a broad one, so 'H2BC21 is more essential than KMT2A' is true of the statistic and misleading about the biology.
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
HMGA1, the one queue member expected to be worth an entry on biological grounds, is a broadly required gene rather than a selective one: mean gene effect -0.8382 with 90.9% of 1,258 lines below -0.5. The remaining four queue members are non-dependencies - NUCKS1 -0.0091 (1.4% of lines below -0.5), PIH1D1 -0.0748 (1.4%), ASB7 -0.0457 (0.5%) - and CD226 has a POSITIVE mean gene effect of +0.1691 with no line below -0.5. Not one of the eight members of this read queue shows a selective dependency pattern.
What this evidence cannot show
'Not one shows a selective dependency pattern' overstates what was measured: this run computed PAN-PANEL means and the fraction below -0.5, and did NOT split KMT2A-rearranged lines from the rest. A selective dependency confined to a handful of index-fusion lines would be invisible in these statistics, so the honest claim is that no member shows a LARGE pan-panel effect that is also distinguishable from class membership - not that selectivity was tested and absent. Rule 84 applies: no median-split proportion is reported. Release identifier unavailable.
Evidence that argues against this (2)
- HMGA1's high essentiality does not argue against its being a real partner, and reading it that way would repeat a mistake this graph has already made. Pan-essentiality and genuine complex membership are independent - the whole point of the MEN1 comparison is that an important interactor can be a mild dependency and an unimportant one can be a hard dependency. Rejecting HMGA1 for being essential would be the inverse of the error cycle 57 made when it nearly rejected FAM9A and RPS11 on the wrong field.
- CD226's positive gene effect is the strongest single argument against the read-queue threshold that admitted these eight. A gene whose loss slightly HELPS 1,258 cell lines, recovered in two publications by anti-tag co-IP against one bait, is what the two-publication criterion lets through - which suggests the threshold is selecting for curation artefacts rather than for biology.
sharpens rather than overturns: the ten-gene-names ground is correct for H3C1 and extends to H2BC10 (5 names on P62807), but it does NOT reach H2BC21 or either ribosomal protein - so 'the histone/ribosomal class' was one label over three grounds
adds two adjudications on cycle 57's own grounds (H2BC10 on ambiguity, RPL13 on pan-essentiality) and shows RPS11 never satisfied the ambiguity ground at all - its accession carries one gene name
adjudicates 2 of the 8 partners that migration's queue produced, leaves 6 explicitly open, and reports that CD226 - positive gene effect - cleared the two-publication threshold
MEN1 -0.362 at 28.9% of 1,258 lines is a MILDER pan-panel dependency than HMGA1 -0.838 or H2BC21 -0.703; the entry flags that this comparison uses the wrong statistic for a selective dependency and must not be read as an argument against targeting menin
DOT1L -0.147 at 12.1%, the mildest of the three complex anchors in this pull, measured as a within-pull control rather than quoted from another cycle
supplies the localisation-and-function field that analysis said was owed for the six it left open, and closes the queue: three rejected, three named as earning entries