Oncofusion

start / graph / ana-q129-psip1-within-length-decile-percentile-in-the-k562-bioid2-map

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q129-the-never-discussed-partners-of-kmt2a-and-psip1

PLANNED: run the size-matched promiscuity correction for PSIP1 in the K562 BioID2 map, the way it was run for KMT2A - fully specified here so the next cycle executes rather than redesigns it

also called PSIP1 length-decile analysis · size-matched promiscuity, PSIP1

A recipe, written down so somebody can just run it: work out how many baits find each protein in one published survey, group proteins by size, and see whether LEDGF is found more often than other proteins of its size. It needs no new experiments, only public data.

The same thing, for a scientist

**Status: PLANNED, not run.** This entry exists because `ana-kmt2a-proximity-promiscuity-is-explained-by-size-97th-percentile-becomes-91st-among-proteins-its-own-length` performed a size-matched correction for KMT2A and **explicitly did not perform it for any other protein**, while `hyp-q129-psip1-may-be-more-proximity-promiscuous-than-its-size-predicts-where-kmt2a-is-not` turns on that number. **All inputs are public and no wet work is involved** [E1, E2]. The design is written out in full below so that the next cycle runs a specified analysis rather than re-deriving one, which is how this project ends up with two analyses of the same thing using different bin edges.

Go deeper
The procedure, fully specified
  1. Pull the bait-prey matrix. https://www.ebi.ac.uk/intact/ws/interaction/findInteractions/39251607?page=N&pageSize=1000, paginated to exhaustion. Use the bare PMID as free text; the field-scoped form pubid:39251607 returns a clean totalElements: 0 on this endpoint 1. Filter every row on publicationPubmedIdentifier == '39251607'. Expect ~19,380 rows, 50 baits, 2,682 preys - and record the numbers you actually get, because a prior PSICQUIC pull gave 19,403 and a silent difference between endpoints is exactly what rule 93 says to announce rather than harmonise.
  2. Compute baits-per-prey. For each prey, the count of distinct baits recovering it. Reference values to reproduce before trusting the pull: median 5, mean 7.23, maximum 40, KMT2A 25, PSIP1 9, MEN1 3.
  3. Resolve lengths. One UniProt canonical sequence length per prey. Use the accession IntAct gives, not the gene symbol - and resolve nothing from recall. Report how many preys fail to resolve and name them, per rule 78: a join that silently drops 5% of preys shifts every decile boundary.
  4. Bin into deciles by length over the preys that resolved, and record the bin edges and the membership count of each bin in the output. KMT2A's published bin is 1,462-5,890 aa with 268 members and a within-bin median of 9 baits.
  5. Report, for PSIP1 and for MEN1 as well as KMT2A: the bin, the bin's member count, the bin's median baits-per-prey, the protein's count, and its within-bin percentile.
  6. Also run the binning-free version, because 90.7 against a bar of 95 is close and deciles depend on edges somebody chose: a regression of baits-per-prey on log10(length) over all preys, reporting each protein's residual and the Spearman rho. If rho is below 0.2, length does not predict recovery in this dataset and both the KMT2A conclusion and the PSIP1 hypothesis are malformed rather than resolved - say so rather than reporting a percentile.
  7. Persist the identities, not just the counts (rule 90): write out the per-prey table with accession, symbol, length, bait count, decile and within-decile rank. The KMT2A analysis's own reproducibility depends on nobody having to re-derive this twice.
What each outcome licenses
resultconsequence
PSIP1's within-decile percentile > 90.7The hypothesis stands. PSIP1's proximity partners cannot be adjudicated as a class on the KMT2A precedent, and each of its nine - including the UTP3 row Q129 declined - needs individual treatment.
PSIP1's within-decile percentile < 90.7The hypothesis is refuted as stated. The class-level treatment extends cleanly from KMT2A to PSIP1 and Q129's UTP3 decline is reinforced.
Spearman rho on log length < 0.2Length does not predict recovery here. Both the KMT2A size explanation and this hypothesis are malformed, and the question moves to abundance - PaxDb or a K562-specific proteome - which the source entry's own open questions already name as the only route left to rescuing the shared-compartment reading.
PSIP1's decile has < 50 membersPercentile unstable. Report uninterpretable and use the regression residual instead.

Note the self-undermining case is the interesting one. A result that weakens Q129's own UTP3 decline is a real possibility and is written into the table rather than left out, because an analysis whose every outcome supports the beat that commissioned it is not an analysis.

Facts and figures
scriptanalyses/psip1-size-matched-proximity-promiscuity/run.py - NOT YET WRITTEN. The full procedure is specified in the 'The procedure, fully specified' section of this entry.
statusplanned
result summaryNOT RUN. Reproduce baits-per-prey (expect median 5, mean 7.23, max 40, KMT2A 25, PSIP1 9, MEN1 3), bin preys by UniProt length into deciles, report PSIP1's and MEN1's within-decile percentile against KMT2A's published 90.7, and run a binning-free regression of baits-per-prey on log10(length) reporting Spearman rho and each protein's residual.
tests hypothesishyp-q129-psip1-may-be-more-proximity-promiscuous-than-its-size-predicts-where-kmt2a-is-not
costcompute only; no wet work, no new data, no restricted access
absence classmeasured-not-analysed
absence class noteThe data exist, are public and are already partly in this project's hands - the 19,380-row pull was made on 2026-09-23 and sits in the session scratchpad. Nothing but compute stands between this entry and a result, which is the definition of measured-not-analysed and the reason this is written as a specified plan rather than as an open question.
inputs
[
  "IntAct REST findInteractions, bare-PMID free-text query on 39251607 (NOT pubid:39251607, which returns 0)",
  "UniProt canonical sequence length for every prey accession",
  "reference values from ana-kmt2a-proximity-promiscuity-is-explained-by-size-97th-percentile-becomes-91st-among-proteins-its-own-length"
]
must report
[
  "the row count actually returned, against 19,380 (REST 2026-09-23) and 19,403 (PSICQUIC, earlier)",
  "the number of preys whose accession failed to resolve to a UniProt length, and their identities (rule 78)",
  "the decile bin edges and per-bin member counts",
  "the per-prey table with accession, symbol, length, bait count, decile and within-decile rank (rule 90: persist the identities, not the count)",
  "the binning-free regression result alongside the decile result, and any disagreement between them"
]
How this could bite us back
  • A planned analysis that nobody runs is worse than no entry, because it looks like coverage. If this is still status 'planned' two cycles from now, the honest move is to delete it or to state in the ledger that the question was raised and abandoned - this project already measured that it states 624 falsifiers and records the outcome of 17.
  • The result changes how much weight this project puts on proximity-derived PSIP1 partners, and PSIP1 is the protein a LEDGIN would act on. Getting the null wrong in either direction misstates the reach of that intervention.
Still open
  1. Should the same correction be run for MEN1 (610 aa, 3 of 50 baits)? The KMT2A entry flags this as a strange omission in a menin-focused project: a menin labelled LESS than its size predicts would be a positive statement about a compact, selectively localised protein rather than a null result.
  2. Should the analysis run on IntAct's curated pair list or on the raw PXD041608 deposition? The curated list has already passed the authors' filter, and if that filter is length-biased the correction partly measures the filter.
  3. Is abundance the better covariate? PaxDb or a K562-specific proteome would let length and abundance be held together, which the KMT2A entry names as the only route left to rescuing the shared-compartment reading for either protein.
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    IntAct REST serves 19,380 rows for PMID 39251607 via the bare-PMID free-text form of findInteractions, resolving to 50 distinct baits and 2,682 distinct preys with median 5 baits per prey, mean 7.23 and maximum 40; PSIP1 is recovered by 9 baits and KMT2A by 25. The field-scoped form 'pubid:39251607' on the same endpoint returns totalElements 0.

    What this evidence cannot show

    Counts differ by 23 rows from the 19,403 recorded in ana-rbp-bioid2-proximity-map-k562 via PSICQUIC on an earlier date; bait count, prey count, median, mean and maximum reproduce exactly. IntAct curates the authors' filtered bona fide pair list, so any percentile computed on it is a percentile of that filter, not of the raw BioID2 data in PXD041608.

    Evidence that argues against this (1)
    • Running the analysis on IntAct's curated subset rather than on the authors' Data file S6 or the raw PXD041608 deposition builds in the filter's behaviour. If the filter is length-biased - and a score built on peptide counts plausibly is - then a length-matched correction on the filtered data could partly be measuring the filter. The procedure above does not control for this and should say so in its own output. Khoroshkin M et al. 2024
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    The reference computation to reproduce and extend is recorded in this graph: KMT2A, 3,969 aa, recovered by 25 of 50 baits, sits at the 97.1st percentile of all preys and the 90.7th within its own 9th length decile (1,462-5,890 aa, 268 members), in which the median prey is recovered by 9 baits against an overall median of 5. UniProt canonical lengths for the two proteins of interest are PSIP1 (O75475) 530 aa and MEN1 (O00255) 610 aa.

    What this evidence cannot show

    The KMT2A figures are quoted from an internal entry and were NOT recomputed in this beat, so they inherit that entry's stated limitations: the deciles were run once with bin edges its author chose, and the entry's own open questions note that length may be the wrong scale because BioID2 labels within a radius and KMT2A is substantially disordered. PSIP1's and MEN1's lengths were fetched live from UniProt this session, not taken from memory or a sibling entry.

    Evidence that argues against this (1)
    • If the KMT2A within-decile figure of 90.7 is itself sensitive to bin edges, then using it as the comparison bar makes this analysis's verdict sensitive to a choice made in a different cycle. That is why step 6 specifies the binning-free regression, and a disagreement between the two forms should be reported rather than resolved in favour of whichever is convenient. Oncofusion Knowledge Graph (internal) 2026
Connected to