The Fanconi signal is real, and it sits in the ubiquitination module
also called FA enrichment test · Fanconi pathway enrichment · FA core complex selectivity
The Fanconi repair pathway has kept turning up in this project without anyone testing whether that was a coincidence. It is not. And the part of the pathway responsible is a specific chemical-tagging step, not the DNA-repair machinery people usually think of.
The same thing, for a scientist
Formal enrichment of the Reactome Fanconi Anemia Pathway (36 of 38 members in scan) among the 17,728-gene KMT2A-rearranged selectivity scan: **median difference −0.0975 against −0.0023 for the rest, Mann-Whitney p = 5.6 × 10⁻⁸**, with **6 members in the top 100 against 0.20 expected** and **0 of 2,000 size-matched random sets** reaching the same median. The enrichment is carried by the **FA core complex and the ID2 monoubiquitination module** — FANCF 24, FANCL 35, UBE2T 51, FANCI 52, FANCA 56, FANCG 75, FAAP100 125, USP1 129 — while **ATR, ATRIP, RPA1/2/3 and the MUS81/EME nucleases do not rank**.
The test, and the controls that make it mean something
| gene set | n in scan | median diff | vs rest | Mann-Whitney p | top-100 obs / exp |
|---|---|---|---|---|---|
| Fanconi Anemia Pathway | 36 | -0.0975 | -0.0023 | 5.6e-08 | 6 / 0.20 |
| Signaling by NOTCH | 188 | 0.0116 | -0.0024 | 0.996 | 3 / 1.06 |
| Cholesterol biosynthesis | 27 | 0.0098 | -0.0024 | 0.948 | 0 / 0.15 |
Cholesterol biosynthesis is the null control and behaves as one — no rank signal, no tail signal, nothing.
NOTCH is the instructive one. Its rank test says the pathway is less selective than background (p = 0.996), yet its top-250 tail is significant (9 observed against 2.65, p = 0.0014). A few NOTCH members are selective; the pathway is not. That is exactly the shape the FA observation had before this test — two members in a top ten — and it is why the observation needed a test rather than a write-up. FA passes both; NOTCH passes only one.
The random control is the strongest line in the table. Of 2,000 size-matched random 36-gene sets drawn from the same 17,728, none reached a median as selective as the FA set.
Which part of the pathway, and why that is the finding
Ranks out of 17,728, most selective first:
| rank | gene | role |
|---|---|---|
| 24 | FANCF | core complex |
| 35 | FANCL | E3 ubiquitin ligase |
| 51 | UBE2T | E2 conjugating enzyme |
| 52 | FANCI | ID2 substrate |
| 56 | FANCA | core complex |
| 75 | FANCG | core complex |
| 125 | FAAP100 | core complex |
| 129 | USP1 | deubiquitinase |
| 211 | SLX4 | nuclease scaffold |
| 295 | FAAP24 | core complex |
| 351 | FANCC | core complex |
| 465 | FANCD2 | ID2 substrate |
The monoubiquitination module is intact at the top: the E2 (UBE2T), the E3 (FANCL), both substrates (FANCI, FANCD2), the deubiquitinase that reverses it (USP1), and the core complex that positions them. What does not rank is the replication-stress arm — ATR, ATRIP, RPA1, RPA2, RPA3 — or the nucleases MUS81, EME1, EME2, ERCC4.
That matters because this project already tested the replication-stress explanation by a completely different route and rejected it: ana-fa-aldehyde-vs-replication found ATR at −0.289 and CHEK1 at −0.317, weakly negative rather than positive, and the aldehyde arm null at ADH5 0.106. Two independent analyses now agree that the FA signal in KMT2A-rearranged lines is not replication stress and not aldehyde detoxification — and this one says where it is instead.
- Test the sub-modules formally: FA core plus ID2 ubiquitination against the ATR/RPA replication-stress arm, within the pathway. The split is currently read off a rank list.
- Does the enrichment survive the lineage split? Seven of the nine panel patients are myeloid, and the earlier FA correlation work found its signal was myeloid-carried with the lymphoid arm non-significant - the wrong lineage for this project.
- Find a second dataset. Three arrivals through one DepMap matrix is one dataset, not three. A published CRISPR screen in KMT2A-rearranged cells from any other source would settle whether this is biology or a property of this panel.
- If the ID2 ubiquitination module is the unit, USP1 is the druggable member - inhibitors exist and are in clinical development for BRCA-mutant cancers. Whether they have ever been tested in KMT2A-rearranged leukaemia is a bounded literature question nobody here has asked.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
The Reactome Fanconi Anemia Pathway is enriched among genes selectively required in KMT2A-rearranged leukemia lines: 36 of 38 members present in a 17,728-gene scan, median selectivity difference -0.0975 against -0.0023 for all other genes, Mann-Whitney one-sided p = 5.6e-08, with 6 members in the top 100 against 0.20 expected, 10 in the top 250 against 0.51, and 14 in the top 500 against 1.02. Of 2,000 size-matched random gene sets, none reached a median as selective.
Fanconi pathway enrichment among KMT2A-rearranged selective dependencies, 17,728-gene scan against Reactome R-HSA-6783310 · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Run 2026-09-20. The gene set was fetched live from Reactome rather than assembled here, so it is defined by a third party. **Selectivity comes from 11 DepMap lines that are nine patients, seven of them myeloid and none an infant**, so this is an enrichment in that panel. The set includes the ubiquitin genes UBB, UBC, UBA52 and RPS27A, which are near-pan-essential and whose ranks should not be read as FA-specific - RPS27A at 214 is plausibly a ribosomal rather than a repair signal. **USP1 has since been removed from this reading.** A co-dependency test across 1,208 lines finds USP1 correlates with the module at r = +0.098 to +0.207 against a module internal median of **0.576**, and with ID1 at −0.022 — it is in neither home. The deubiquitinase should not be counted among the module members carrying this enrichment. The enrichment statistic is unaffected, because it was computed over the whole Reactome pathway rather than the hand-picked subset. See `ana-usp1-is-in-neither-home`.
Evidence that argues against this (3)
- The instruments differ but the data do not. This test and the pathway-correlation work that first found the FA signal both draw on the same DepMap Chronos matrix and the same eleven-line panel, so a panel-level artefact - a shared lineage bias, a shared screen batch - would produce the signal in both. Three arrivals by three analyses of one dataset is weaker than three arrivals by three datasets. SIB Swiss Institute of Bioinformatics 2026
- The NOTCH control shows the tail test alone is not reliable, and the FA result leans partly on tail statistics. NOTCH has 9 of 250 against 2.65 expected (p = 0.0014) while being LESS selective than background overall. A hypergeometric tail p is sensitive to a handful of genes, and for FA those genes are the core complex - which is the interesting reading, but also the one a sceptic would call a small-n coincidence dressed as a pathway. Oncofusion knowledge graph (internal analysis) 2026
- Tested directly rather than argued: USP1's cross-line dependency profile matches neither the FA module nor ID1, and the negative control ACTB (-0.127) exceeds several of the module correlations in magnitude. Correlations of that size in this matrix are what unrelated genes look like. Oncofusion knowledge graph (internal analysis) 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
The enrichment is carried by the FA core complex and the ID2 monoubiquitination module rather than by the replication-stress arm. The most selective members are FANCF (rank 24), FANCL (35, the E3 ligase), UBE2T (51, the E2), FANCI (52), FANCA (56), FANCG (75), FAAP100 (125) and USP1 (129, the deubiquitinase), with FANCD2 at 465; ATR, ATRIP, RPA1, RPA2, RPA3, MUS81, EME1 and EME2 do not rank comparably.
Fanconi pathway enrichment among KMT2A-rearranged selective dependencies, 17,728-gene scan against Reactome R-HSA-6783310 · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Ranks are out of 17,728 protein-coding genes on a single selectivity statistic, and no formal test was run comparing the ubiquitination sub-module against the replication-stress sub-module - the split is read off the rank list. A sub-pathway-level test would be the stricter version and has not been done.
Evidence that argues against this (2)
- A ranking is not a mechanism. That the E2, E3, substrates and deubiquitinase all rank high is consistent with the module being required, and equally consistent with all of them being required for the same upstream reason that has nothing to do with KMT2A - for instance if these lines carry replication or crosslink stress from some other source. The test establishes selectivity relative to other leukaemia lines, not causation by the fusion. Reactome 2026
- The FA pathway has now survived exclusion of both rival mechanisms and gained a formal enrichment, and it still has no mechanism connecting it to KMT2A::AFF1. Accumulating statistical support for a link whose biology nobody can state is a pattern this project should be suspicious of rather than reassured by. Oncofusion knowledge graph (internal analysis) 2026
independent complex-level confirmation of the FA core complex signal