Oncofusion

start / graph / hyp-q129-psip1-may-be-more-proximity-promiscuous-than-its-size-predicts-where-kmt2a-is-not

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q129-the-never-discussed-partners-of-kmt2a-and-psip1

The size correction that dissolved KMT2A's proximity promiscuity should run the OTHER WAY for PSIP1: at 530 aa, recovery by 9 of 50 BioID2 baits may be MORE than its length predicts, which would make PSIP1 rather than KMT2A the promiscuous prey in this project's core pair

also called PSIP1 length-decile hypothesis · LEDGF proximity promiscuity

A survey found the big protein this project studies near half of fifty tagged baits, and a previous analysis showed that is mostly because it is enormous and easy to bump into. Its much smaller partner LEDGF was found near nine baits, which for a protein that size may actually be more than expected.

The same thing, for a scientist

**Statement.** In the Khoroshkin K562 50-bait BioID2 map, PSIP1 (530 aa) is recovered by **9 of 50 baits**, the 68th percentile of the 2,682 preys [E1]. This graph has already shown that KMT2A's far more striking **25 of 50** collapses from the 97.1st percentile of all preys to the **90.7th** within its own 1,462-5,890 aa length decile, where the median prey is recovered by **9 baits rather than the overall median of 5** [E2]. **Nobody has run that correction for PSIP1, and the arithmetic points the other way.** If large proteins in this map are recovered by roughly twice as many baits as typical ones, then the expected count scales with length - and a 530 aa protein achieving 9, the figure that is merely *median* for the largest decile, may sit high within its own. **The hypothesis is that PSIP1's within-decile percentile exceeds KMT2A's**, i.e. that once size is held constant the promiscuous prey in this project's central pair is the tether, not the methyltransferase. **This matters beyond bookkeeping.** `ana-kmt2a-proximity-promiscuity-is-explained-by-size...` used the size result to conclude that this census **cannot support individual proximity partners for KMT2A** [E2]. That conclusion was applied to KMT2A. **If PSIP1's promiscuity is NOT explained by size, the same logic does not automatically extend to PSIP1's nine partners, and they would need adjudicating individually rather than as a class** - which includes the UTP3 row this beat declined [E3]. The hypothesis is therefore self-undermining in a useful way: confirming it weakens this beat's own strongest quantitative decline.

Go deeper
Why this is stated as a hypothesis and not computed here

The computation needs the length of every one of the 2,682 preys in the map, binned into deciles, with PSIP1's rank read within its own bin - the same procedure the KMT2A analysis ran. That means 2,682 UniProt length lookups or one bulk download, which was outside this beat's scope after thirteen partner adjudications.

It is cheap, it needs no wet work, and it is fully specified in ana-q129-psip1-within-length-decile-percentile-in-the-k562-bioid2-map, written as a planned analysis so the next cycle can run it without re-deriving the design.

The prior is genuinely uncertain, which is why it is set at 0.5. Two readings are about equally available. Recovery may scale with length because a larger protein presents more accessible lysines within a labelling radius - in which case PSIP1's 9 is unremarkable for a small protein only if the scaling is weak. Or recovery may scale with abundance and residence time rather than length, in which case the length decile is the wrong control for both proteins and the KMT2A result needs its own re-examination, which that entry's own open questions already raise.

Facts and figures
statementIn the Khoroshkin K562 50-bait BioID2 map, PSIP1's baits-per-prey percentile computed WITHIN its own protein-length decile is higher than KMT2A's within-decile percentile of 90.7 - i.e. once size is held constant, PSIP1 rather than KMT2A is the promiscuous prey in this project's central pair.
origindata-pattern
prior probability0.5
absence classmeasured-not-analysed
absence class noteEvery number needed is already public: the bait-prey matrix is in IntAct and in Data file S6 of PMID 39251607, and the lengths are in UniProt. This is compute with no wet work and no new data, which is why it is measured-not-analysed. It was not run in this beat because thirteen partner adjudications consumed the budget, and that is a scheduling fact rather than an obstacle.
predictions
[
  "PSIP1 (530 aa) falls in a length decile whose median baits-per-prey is at or below the overall median of 5, so its 9 baits sit well above that decile's median.",
  "PSIP1's within-decile percentile exceeds 90.7.",
  "Across all 2,682 preys, baits-per-prey correlates positively with log protein length, with a Spearman rho above 0.2."
]
experiments
[
  "exp-q129-a-kmt2a-bait-proximity-experiment-in-a-kmt2a-rearranged-leukaemia-line"
]
analyses
[
  "ana-q129-psip1-within-length-decile-percentile-in-the-k562-bioid2-map"
]
falsifiers
[
  "SCOREABLE: In the Khoroshkin K562 BioID2 map (PMID 39251607, all 2,682 preys, deciles by UniProt canonical sequence length), PSIP1's within-decile baits-per-prey percentile is BELOW 90.7 - KMT2A's published within-decile figure. If PSIP1 comes in below that, the hypothesis is refuted as stated and the class-level treatment of proximity partners extends cleanly from KMT2A to PSIP1.",
  "SCOREABLE: Across all 2,682 preys in that map, Spearman correlation between baits-per-prey and log10(UniProt length) is BELOW 0.2. That would mean length does not predict recovery in this dataset at all, which makes both this hypothesis and the KMT2A size explanation malformed rather than false, and redirects the question to abundance (PaxDb or a K562-specific proteome) as the source entry's own open questions already propose.",
  "SCOREABLE: PSIP1's within-decile percentile exceeds 90.7 but the decile itself has fewer than 50 members, making the percentile unstable. In that case the result is recorded as uninterpretable rather than as support, and a continuous regression of baits-per-prey on log length with PSIP1's residual reported replaces the binning - which is the form that does not depend on bin edges anyone chose."
]
How this could bite us back
  • If PSIP1 is the promiscuous node, then proximity-derived PSIP1 partners across this graph - not only the UTP3 row declined in this beat - need the same discount currently applied to KMT2A's, and PSIP1 is the protein a LEDGIN would act on. Over-crediting its neighbourhood would inflate the apparent reach of that intervention.
  • The opposite risk is real too. If PSIP1's breadth is biological - a tether that genuinely sits near many transcription-associated proteins - then discounting it as promiscuity would delete the very property that makes it the menin-KMT2A complex's chromatin address, which is the thing this project cares about most.
Still open
  1. Is length the right covariate at all for a proximity-labelling radius, or should it be accessible surface, disorder content, or abundance? KMT2A is substantially disordered and PSIP1 has a long disordered region too, so residue count may overstate expected labelling for both.
  2. If PSIP1 IS more promiscuous than its size predicts, does that have a biological reading rather than a technical one? LEDGF is a chromatin tether that binds H3K36me2/3 nucleosomes broadly across gene bodies, so being near many transcription-associated proteins could be what the protein does rather than an artefact of the assay.
  3. Does the same within-decile correction change the verdict for MEN1, recovered by 3 of 50 at 615 aa? The source entry flags this as a strange omission in a menin-focused project, and a menin that is labelled LESS than its size predicts would be a positive statement about a compact, selectively localised protein rather than a null.
Every source behind this entry (3)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    In the Khoroshkin 50-RBP BioID2 map in K562 (PMID 39251607), IntAct REST serves 19,380 rows resolving to 50 baits and 2,682 distinct preys, median 5 baits per prey, mean 7.23, maximum 40. PSIP1 (O75475, 530 aa) is recovered by 9 baits - AGGF1, FASTKD2, HNRNPC, ILF3, PPIL4, RBM15, SBDS, U2AF1, UTP3 - which is the 68th percentile, since 851 of 2,682 preys are recovered by 9 or more baits. KMT2A (Q03164, 3,969 aa) is recovered by 25 baits, the top 2.9%.

    What this evidence cannot show

    Counted from IntAct's curation of the deposition, not from the authors' Data file S6. The row count differs by 23 from the 19,403 recorded in ana-rbp-bioid2-proximity-map-k562, which used PSICQUIC on an earlier date; bait count, prey count, median, mean and maximum reproduce exactly. K562 carries BCR::ABL1 and no KMT2A rearrangement, so the whole dataset is a wild-type control for this project's disease.

    Evidence that argues against this (1)
    • IntAct curates the authors' bona fide pair list, which has already passed their filtering. The raw unfiltered hit lists in PXD041608 would give different degrees for both proteins, and ana-rbp-bioid2-proximity-map-k562 already asks whether the raw lists are richer than the curated subset. Percentiles computed on a filtered list are percentiles of that filter. Khoroshkin M et al. 2024
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    This graph's entry ana-kmt2a-proximity-promiscuity-is-explained-by-size-97th-percentile-becomes-91st-among-proteins-its-own-length records that KMT2A's recovery by 25 of 50 baits in the K562 BioID2 map places it at the 97.1st percentile of all preys but only the 90.7th among the 268 proteins in its own length decile (1,462-5,890 aa), where the median prey is recovered by 9 baits against an overall median of 5. It concludes that the census cannot support individual proximity partners for KMT2A and that the RNA-processing block should be adjudicated as a class or not at all. The equivalent within-decile computation was NOT performed for PSIP1 in that entry.

    What this evidence cannot show

    An internal entry, cited to establish what this project has already computed rather than as independent evidence; its underlying computation was not re-run here. That entry's own open questions flag two limitations this hypothesis inherits: the decile binning was run once with bin edges the author chose, and length may be the wrong scale altogether because BioID2 labels within a radius, so accessible surface and residence time are what should scale rather than residue count - and KMT2A is substantially disordered.

    Evidence that argues against this (1)
    • If length is the wrong scale, this hypothesis is malformed rather than false: comparing PSIP1 to length-matched peers would then be no more meaningful than comparing KMT2A to them, and the correct control would be abundance in K562 from PaxDb or a line-specific proteome. That alternative control was named in the source entry's open questions and has not been run for either protein. Oncofusion Knowledge Graph (internal) 2026
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Q129 declined the PSIP1-UTP3 row partly on the ground that PSIP1's nine recovering baits contain only two nucleolar proteins against a panel base rate of nine in fifty (exact binomial p = 0.50). That argument treats PSIP1's nine baits as an unremarkable neighbourhood. If PSIP1 turns out to be more promiscuous than its length predicts, the neighbourhood is instead unusually large for a small protein, which strengthens the background reading for UTP3 rather than weakening it - but it simultaneously removes the licence to adjudicate PSIP1's proximity partners as a class rather than individually.

    What this evidence cannot show

    This is a self-citation to an entry written in the same beat by the same agent, which is the weakest form of internal citation and is used here only to state what that entry's argument was, not to support it. The two consequences named - strengthening the background reading while removing the class-adjudication licence - pull in opposite directions and have not been reconciled; which dominates depends on the size of the effect, which is exactly what has not been measured.

    Evidence that argues against this (1)
    • The base-rate test and the length test measure different things - composition of the neighbourhood versus its size - and a result on one does not automatically bear on the other. Treating them as coupled may be an error of this hypothesis rather than a real interaction between the two analyses. IntAct (EMBL-EBI) 2026
Connected to