Oncofusion

start / graph / ana-the-interactomes-quality-frontier-is-already-fully-written

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

All 56 core-complex partners that anyone deliberately fished for already have an entry: the unwritten interactome is 100% by-product

also called bait-supported worklist · interactome frontier · Q5 gap reframed · interactome coverage

This project carried a figure saying only about a third of the fusion's known partners were written up, which sounded like a large research gap. Splitting them by whose experiment found them changes the picture completely: every partner somebody deliberately went looking for is already written, and everything still missing turned up incidentally in somebody else's experiment.

The same thing, for a scientist

The coverage ledger records the core-complex interactome at **31.8% one-entry-per-thing coverage with 136 partners absent entirely**, and lists a curated read queue as the next beat. Grading partners by experimental role instead: - **56 partners of KMT2A, MEN1, PSIP1, AFF1, AFF4 or DOT1L are bait-supported** — at least one curated IntAct interaction exists in which the core-complex protein was the **bait**. - **0 of those 56 lack an entry in this graph.** The denominator is the point: 56, not zero, so the instrument works — this is not the `FULL_TEXT:` failure where a query returned nothing because it did not compose. **So the 136 absent partners are entirely prey-only**, consistent with the finding that 67.9% of all partner-bait pairs rest on rows where our protein was prey. **The unwritten interactome is 100% by-product evidence**, and the read queue this ledger was about to dispatch a beat against is drawn from it.

Go deeper
Why the 31.8% figure was misleading and what to say instead

'A third of documented interactors have an entry' invites the reading that two-thirds of the real biology is unwritten. The honest restatement is:

100% of the directed-evidence interactome is written. 0% of the by-product interactome is.

Those are different claims with different consequences. The first implies a large reading backlog. The second says the backlog consists of proteins that appeared in somebody else's pulldown, in somebody else's cell line, chosen to study somebody else's question — and that writing them up would add entries whose evidence line must read "detected as prey in an atlas of RNA-binding proteins in K562".

That is not worthless and this entry does not say it is. Some of those partners are real. But it is a different kind of work from what the ledger implied, and it should be chosen deliberately rather than inherited from a coverage percentage.

The practical consequence: I did not dispatch an interactome beat this cycle. The high-quality work is done, and the remaining work is low-grade by construction. A beat there would spend a research agent writing candidate, one method, one lab, wrong cell type twenty times over.

What this says about where the real gap is

The interactome's problem is not coverage, it is that almost nobody has fished with these proteins. Fifty-six bait-supported partners across six proteins is a small number for a complex this important, and the companion analysis showed the distribution is lopsided: PSIP1/LEDGF has one bait-supported partner out of thirty-six, while DOT1L — with a drug programme behind it — has fourteen of twenty-six.

So the binding constraint is experimental, not curatorial. The graph cannot write entries for directed evidence that does not exist. That is why the proposed proximity-labelling experiment matters more than another reading pass: a TurboID of KMT2A::AFF1 in an index-fusion line would move the bait-supported count for the first time in years, rather than redistributing what is already catalogued.

And it is a measurement the PI can act on. 'Map every interactor' is currently limited by how few interactors anyone has looked for, not by how many this project has transcribed.

Facts and figures
n bait supported partners56
n missing an entry0
companion67.9% of all partner-bait pairs are prey-only
ledger figure being reframed
{
  "coverage_pct": 31.8,
  "absent_entirely": 136
}
How this could bite us back
  • Do not quote '100% of the directed interactome is written' as though the interactome were finished. It means the graph has kept up with a small and lopsided body of directed work - 56 partners across six proteins - not that 56 is the right number. The constraint is that few people have fished with these proteins.
  • And the zero flatters this project partly by construction: the interactome beats were built from IntAct, so the best-curated partners are the ones earlier beats were most likely to find. It shows thorough reading of one source, not completeness.
Still open
  1. Does the same hold for the 3'-partner private interactomes, which the ledger calls the emptiest measured region? Only the six core proteins were graded.
  2. Would a partner documented in a paper IntAct has not curated be invisible to this whole line of analysis? Almost certainly, and nothing here bounds how many there are.
  3. Is PSIP1's low bait-supported count neglect, or a literature that used structural methods which generate no bait rows?
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis

    Of 56 partners of KMT2A, MEN1, PSIP1, AFF1, AFF4 or DOT1L that are supported by at least one curated IntAct interaction in which the core-complex protein was the experimental bait, zero lack a gene entry in this knowledge graph, so the 136 partners recorded as absent from the graph are entirely prey-only or role-unrecorded.

    What this evidence cannot show

    Run 2026-09-20. **Bait-supported is a floor** - one curated bait row scores the same as twenty, and a bait row from a large two-hybrid screen counts. **Matching against the graph is by `gene-<symbol>` id and `data.gene_symbol`**, so a partner recorded here under a different symbol or an isoform accession could be wrongly counted as present; the error direction inflates coverage. IntAct curates what was published and pools screens of very different quality. **The denominator is reported because a zero without one is worthless** - 56 partners were found, so the query composes.

    Evidence that argues against this (1)
    • A zero here is partly a compliment this project pays itself. The graph's interactome beats were built largely FROM IntAct, so the proteins with the best-curated interaction records are exactly the ones earlier beats were most likely to have found and written. The result shows the reading was thorough against that source; it does not show the source is complete, and a partner documented only in a paper IntAct has not curated would be invisible to both. EMBL-EBI IntAct 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    The distribution of bait-supported partners across the six core-complex proteins is lopsided, with PSIP1 carrying 1 bait-supported partner of 36 and DOT1L 14 of 26, indicating that directed interaction work on these proteins is concentrated on the one with a drug-development programme behind it.

    What this evidence cannot show

    Attributing the spread to commercial attention is an interpretation, not a measurement - a protein studied intensively by structural or genetic methods would also score poorly here, and LEDGF has a large structural literature from HIV integrase work that generates few IntAct bait rows. **Moderate for that reason.** Counts are partner-bait pairs from one database on one day.

    Evidence that argues against this (1)
    • LEDGF's low count may reflect that its best-characterised interactions are with HIV integrase and with the MLL complex, both studied structurally rather than by pulldown, rather than that nobody has looked. A low bait-supported count is evidence about the method used, not about how well understood a protein is. EMBL-EBI IntAct 2026
Connected to