The MRN asymmetry adjudicated: RAD50 -0.114 sits with NBN -0.047 and not with MRE11 -0.542, so it is not one subunit against two - and yet MRE11 and RAD50 covary across 1,208 cell lines at r = 0.554, the same as BRCA1 against BARD1, against a background of zero
also called the MRN eleven-fold puzzle, adjudicated · RAD50 settles the MRE11 versus NBN gap · covariation against per-line dependency calls
Two entries in this graph noticed that one member of a three-protein machine looks eleven times more important than another in the big knockout screens, and neither could explain it. Fetching the third member shows the machine's parts rise and fall together across a thousand cell lines exactly as a true partnership should, while their average scores are simply on different scales - so the puzzle was about the yardstick, not about the biology.
The same thing, for a scientist
**The puzzle, as `gene-mre11` and `gene-nbn` both state it.** One Chronos call, one Open Targets release, near-identical denominators: **MRE11 -0.542 with 56.1% of 1,258 lines below -0.5, NBN -0.047 with 1.3%.** Two subunits of an obligate complex, eleven-fold apart, and NBN is the one whose biallelic loss causes Nijmegen breakage syndrome. Both entries name **RAD50** as the unfetched discriminator. **Step 1 - fetch RAD50, and the simplest explanations die** [E1]. RAD50 is **-0.1141 over 1,257 screens, 5.2% below -0.5.** It sits with NBN. | | MRE11 | RAD50 | NBN | |---|---:|---:|---:| | n screens | 1,258 | 1,257 | 1,258 | | mean Chronos | **-0.5423** | **-0.1141** | **-0.0471** | | % below -0.5 | 56.1% | 5.2% | 1.3% | | haem/lymphoid (n=141) | -0.586 | -0.240 | +0.003 | So it is **MRE11 alone**, not one-against-two. That rules out *"the nuclease matters and the scaffold does not"* - RAD50's ATPase is required for MRE11's endonuclease in every reconstitution - and it rules out *"NBN is the odd one out"*, which is how the two existing entries framed it. **Step 2 - ask whether they move together, which is a different question from whether their means match** [E2]. Across the 1,208 lines screened for all of them, with a pre-named obligate-heterodimer positive control and four unrelated-pair negative controls run through the identical code path: | pair | Pearson r | 95% CI | Spearman | |---|---:|---|---:| | **MRE11 ~ RAD50** | **0.554** | **0.513-0.592** | 0.551 | | **BRCA1 ~ BARD1** (positive control) | **0.560** | **0.520-0.598** | 0.549 | | MRE11 ~ NBN | 0.321 | 0.269-0.371 | 0.336 | | RAD50 ~ NBN | 0.283 | 0.231-0.334 | 0.290 | | MRE11 ~ ATM | 0.321 | 0.270-0.371 | 0.341 | | CBX3 ~ CBX5 | 0.170 | 0.114-0.224 | 0.168 | | MRE11 ~ BRCA1 | -0.004 | -0.060-0.053 | 0.010 | | MRE11 ~ MEN1 | -0.004 | -0.060-0.052 | 0.001 | | BRCA1 ~ ATM | -0.018 | -0.075-0.038 | -0.001 | | CBX1 ~ CBX3 | -0.019 | -0.076-0.037 | -0.018 | **MRE11~RAD50 and BRCA1~BARD1 have almost coincident confidence intervals, and the unrelated-pair background is zero.** The complex's dependency structure is intact. **Step 3 - and the per-line call says the exact opposite, correctly** [E3]. Applying the usual -0.5 dependency threshold line by line: **292 of 1,208 lines (24.2%) are MRE11-dependent while both RAD50 and NBN are above -0.2, and exactly ONE line in 1,208 has all three below -0.5.** Anyone asking "in how many lines is the MRN complex a dependency?" gets an answer that says the complex is entirely incoherent, and they are right about the question they asked. **The verdict.** The asymmetry is **a per-gene offset, not a difference in biology**: the three subunits share the same across-line dependency structure and differ in where that structure is centred. That decisively excludes explanation (a), "the complex's members do different biology", which is the reading both existing entries lean toward. It does **not** discriminate between (b) guide behaviour and (c) something in the screen's per-gene scaling - both produce an offset with intact covariation. **Naming what would**: guide-level log-fold-change against the copy number at 11q21 (MRE11), 5q31 (RAD50) and 8q21 (NBN), which is the design already written for CBX1 in `exp-guide-level-decomposition-of-the-cbx1-depmap-screen` and has never been run for MRN. **And the transferable half.** Two readings of one dataset, both correct, giving opposite answers to a question nobody had split in two. That is the same shape as rule 96's cashed falsifier, where H3K9me3 fired at one interval scale and not at another. **A complex is coherent in DepMap if its members covary; whether its members are each called dependencies is a fact about a threshold.**
| script | analyses/q37-mrn-depmap-adjudication/run.py |
|---|---|
| status | run |
| run date | 2026-09-22 |
| run by | Q37-archive-sweep |
| result summary | RAD50 -0.114 groups with NBN -0.047, not MRE11 -0.542, so the asymmetry is MRE11-alone. But MRE11~RAD50 covary at r=0.554 across 1,208 lines, matching the BRCA1~BARD1 positive control at 0.560 against a ~0 background. The gap is an offset in level with intact structure - a scaling or guide-behaviour explanation, not a biology one. |
| endpoint | https://api.platform.opentargets.org/api/v4/graphql - target(ensemblId).depMapEssentiality{ tissueName screens{ depmapId cellLineName diseaseFromSource geneEffect } } |
| symbol resolution | every symbol resolved to its HGNC PRIMARY ensembl_gene_id via rest.genenames.org, keeping only the row whose `symbol` equals the symbol asked for (rules 62, 78), and the Open Targets approvedSymbol read back and asserted equal for all 18 genes |
| an instrument observation that is not explained | the screen count returned by this endpoint varies far more than rule 83's gene-specific-denominator note predicts: 1,258 for most genes, 1,096 for OR5A1, 394 for RPL13A, 80 for DNA2 (mean -2.712, 100% below -0.5), 5 for OR4F5 and 0 for POLR2A. The genes with tiny n are the most extreme dependencies, which is a suspicious pattern and is NOT understood here. No claim in this entry rests on a gene with n < 1,000, and the DNA2 and RPL13A numbers are reported only as instrument observations. |
| relationship to existing entries | Answers the first open question of `gene-mre11` and the first open question of `gene-nbn`, both of which name RAD50 as the unfetched discriminator. It does not overturn either entry's numbers, which are reproduced here to four decimal places from the same release; it reframes what they mean. |
release
{
"source": "Open Targets Platform",
"dataVersion": {
"year": "26",
"month": "06",
"iteration": null
},
"fetched": "2026-09-22"
}per gene — the numbers behind the chart
{
"RAD50": {
"ensembl": "ENSG00000113522",
"n_screens": 1257,
"mean": -0.1141,
"median": -0.09,
"frac_below_minus_0_5": 0.0517,
"min": -1.501,
"max": 0.692
},
"MRE11": {
"ensembl": "ENSG00000020922",
"n_screens": 1258,
"mean": -0.5423,
"median": -0.545,
"frac_below_minus_0_5": 0.5612,
"min": -1.803,
"max": 0.22
},
"NBN": {
"ensembl": "ENSG00000104320",
"n_screens": 1258,
"mean": -0.0471,
"median": -0.037,
"frac_below_minus_0_5": 0.0127,
"min": -0.794,
"max": 0.799
},
"BRCA1": {
"n_screens": 1258,
"mean": -0.4888,
"frac_below_minus_0_5": 0.4404
},
"BARD1": {
"n_screens": 1258,
"mean": -0.5742,
"frac_below_minus_0_5": 0.5819
},
"RBBP8": {
"n_screens": 1258,
"mean": -0.9346,
"frac_below_minus_0_5": 0.9515
},
"ATM": {
"n_screens": 1258,
"mean": -0.0156,
"frac_below_minus_0_5": 0.0111
},
"EXO1": {
"n_screens": 1258,
"mean": -0.1273,
"frac_below_minus_0_5": 0.0429
},
"KDM5A": {
"n_screens": 1258,
"mean": -0.1185,
"frac_below_minus_0_5": 0.0644
},
"MEN1": {
"n_screens": 1258,
"mean": -0.3616,
"frac_below_minus_0_5": 0.2893
}
}haematopoietic and lymphoid subset
{
"n_lines": 141,
"MRE11": {
"mean": -0.586,
"frac_below_minus_0_5": 0.589
},
"RAD50": {
"mean": -0.24,
"frac_below_minus_0_5": 0.184
},
"NBN": {
"mean": 0.003,
"frac_below_minus_0_5": 0.007
},
"BRCA1": {
"mean": -0.286,
"frac_below_minus_0_5": 0.184
},
"BARD1": {
"mean": -0.456,
"frac_below_minus_0_5": 0.426
},
"RBBP8": {
"mean": -0.861,
"frac_below_minus_0_5": 0.922
},
"MEN1": {
"mean": -0.602,
"frac_below_minus_0_5": 0.482
}
}pairwise correlations 1208 shared lines — the numbers behind the chart
{
"MRE11~RAD50": {
"pearson": 0.554,
"ci95": [
0.513,
0.592
],
"spearman": 0.551
},
"BRCA1~BARD1 (POSITIVE CONTROL)": {
"pearson": 0.56,
"ci95": [
0.52,
0.598
],
"spearman": 0.549
},
"MRE11~NBN": {
"pearson": 0.321,
"ci95": [
0.269,
0.371
],
"spearman": 0.336
},
"RAD50~NBN": {
"pearson": 0.283,
"ci95": [
0.231,
0.334
],
"spearman": 0.29
},
"NBN~BRCA1": {
"pearson": 0.24,
"ci95": [
0.186,
0.292
],
"spearman": 0.259
},
"MRE11~ATM": {
"pearson": 0.321,
"ci95": [
0.27,
0.371
],
"spearman": 0.341
},
"CBX3~CBX5": {
"pearson": 0.17,
"ci95": [
0.114,
0.224
],
"spearman": 0.168
},
"MRE11~BRCA1 (negative control)": {
"pearson": -0.004,
"ci95": [
-0.06,
0.053
]
},
"MRE11~MEN1 (negative control)": {
"pearson": -0.004,
"ci95": [
-0.06,
0.052
]
},
"MRE11~RBBP8 (negative control)": {
"pearson": -0.037,
"ci95": [
-0.093,
0.019
]
},
"BRCA1~ATM (negative control)": {
"pearson": -0.018,
"ci95": [
-0.075,
0.038
]
},
"CBX1~CBX3 (negative control)": {
"pearson": -0.019,
"ci95": [
-0.076,
0.037
]
}
}per line dependency calls — the numbers behind the chart
{
"n_lines_with_all_three_screened": 1208,
"MRE11_below_minus_0_5_while_RAD50_and_NBN_above_minus_0_2": 292,
"pct": 24.2,
"all_three_below_minus_0_5": 1,
"pct_all_three": 0.08
}controls in the same call
{
"low_dependency": {
"OR4F5": {
"n_screens": 5,
"mean": -0.0425
},
"OR5A1": {
"n_screens": 1096,
"mean": -0.1433,
"frac_below_minus_0_5": 0.0091
}
},
"pan_essential": {
"RPL13A": {
"n_screens": 394,
"mean": -2.0745,
"frac_below_minus_0_5": 1
}
},
"failed_control_reported_not_dropped": {
"POLR2A": "ENSG00000181222 resolved cleanly through HGNC and Open Targets returned ZERO depMapEssentiality blocks. A pan-essential positive control that returns no data is a failed control and is recorded as one."
},
"why_this_matters": "Q35's first DepMap control returned a null target and demonstrated nothing. Running low-dependency and pan-essential controls in the SAME call is what distinguishes 'RAD50 is not a dependency' from 'the endpoint is not returning dependencies'."
}what is adjudicated and what is not
{
"EXCLUDED": "(a) the complex's members do different biology. Their dependency profiles covary as tightly as an obligate heterodimer's; that is not what different biology looks like.",
"NOT_DISCRIMINATED": "(b) CRISPR guide behaviour, e.g. cutting toxicity that differs by locus copy number, and (c) per-gene scaling inside the screen. Both produce a shift in level with intact structure, which is exactly what is observed.",
"WHAT_WOULD_DISCRIMINATE": "guide-level log-fold-change for MRE11 (11q21), RAD50 (5q31) and NBN (8q21) against each line's copy number at those loci, plus residual protein by western in a few lines. The guide-level design is already written in exp-guide-level-decomposition-of-the-cbx1-depmap-screen and has never been run for MRN."
}- If MRE11's depth is guide toxicity rather than dependency, then MRE11-directed inhibitors (the mirin series) would be far less selective in leukaemia than the Chronos mean implies, and a programme justified by 'MRE11 is a dependency in 56% of lines' would be justified by an artefact.
- The opposite error is also available and is the one this entry could cause: concluding that because the subunits covary, MRN is 'one target' and hitting any subunit will do. RAD50 and NBN are near-zero dependencies in almost every line, so a RAD50-directed agent would have no single-agent effect however coherent the complex is.
- Is the MRE11 offset copy-number cutting toxicity? MRE11 is at 11q21, RAD50 at 5q31 and NBN at 8q21; a guide-level log-fold-change against each line's copy number at its own locus is the measurement, it needs no wet work, and it has never been run for any MRN subunit in this graph.
- How many other 'incoherent complex' observations in this corpus are offsets with intact covariation? The method here is four lines of code and applies to PRC1.1, NELF, FACT, PRC2, the 7SK snRNP and the NSL complex, all of which have subunit entries carrying DepMap numbers and none of which has been checked for covariation.
- Does the covariation hold inside the 141 haematopoietic lines alone? The means were computed for that subset and the correlations were not, and this graph's question is a blood-cell question.
the discriminator both gene-mre11 and gene-nbn named as unfetched
MRE11 is the single offset member: -0.542 against RAD50 -0.114 and NBN -0.047, yet it covaries with RAD50 at r=0.554
NBN is not the odd one out; RAD50 sits with it and the pair's correlation with MRE11 is pathway-level (0.28-0.32), not complex-level
the written design that would discriminate guide behaviour from screen scaling, never run for MRN
the DepMap adjudication this entry was written to close