PSIP1-RBPJ - DECLINED. A 214-purification transcription-factor TAP atlas that this beat's own atlas heuristic failed to flag, and the Notch connection is the reason to say so out loud
also called LEDGF-RBPJ · PSIP1-RBPJ · CSL · p75-RBPJ
The Notch signalling DNA-binding protein RBPJ turned up with LEDGF in a big survey of what sticks to transcription factors. Notch matters enormously in leukaemia, which is exactly why this weak hit needs saying plainly rather than quietly promoting.
The same thing, for a scientist
**Verdict: DECLINED, and filed because the partner is one where a weak row is most likely to be over-read.** IntAct holds one PSIP1-RBPJ record: **TAP**, *association*, HEK293T, PSIP1 as prey, intact-miscore 0.35, from **PMID 25609649** [E1] - Li, Wang, Wang, Malovannaya, Xi, Li, Guerra, Hawke, Qin and Chen, *Mol Syst Biol* 2015, which performed **214 tandem-affinity purifications across 56 transcription factors and 37 Forkhead-box family members** and reported **2,156 high-confidence protein-protein interactions**, concluding that TFs form distinct chromatin-associated and soluble complexes [E2]. **This is a proteome-scale atlas by any reasonable definition and this beat's atlas heuristic did not flag it**, because it supports only five prey-only partners across the six baits and the threshold was >=15. Rule 92 says to print the heuristic's answer and hand-read the top; done, and the heuristic is wrong here. The same publication supplies four other prey-only rows in this census: **FOXB1** and **JUN** and **FOXE1** and **FOXK2**.
The heuristic failure, recorded as a result
This beat flagged a publication ATLAS-LIKE if it supported >=15 distinct prey-only partners across the six baits. Five publications were flagged: BioPlex 3.0 (53), Hein 2015 (33), the K562 RBP BioID map (28), Fasci crosslink-MS (16) and BioPlex 2.0 (15).
PMID 25609649 supports five and was not flagged. It is a 214-purification, 2,156-interaction atlas.
Why the heuristic missed it, and the general lesson: the detector counts partners of our six baits, not the atlas's own size. An atlas that happens to touch our baits rarely looks focused. A partner-count threshold measures the overlap with our interest, not the study's design - and a study with a 2,156-interaction output that intersects our baits five times is exactly as atlas-like as one that intersects them fifty times. The fix is to read the title, which takes two minutes and is what rule 92 prescribes; the heuristic's answer is retained in census.json so the size of its error is recoverable.
Consequence for this beat's ranking: the n_focused_pmids column in census.json overstates focused support for at least five partners - RBPJ, FOXB1, FOXE1, FOXK2 and JUN - and probably for others not hand-read. The ranking is a starting point, not a verdict, and this is the concrete way in which it is wrong.
Why RBPJ specifically needed a written decline
RBPJ (CSL) is the DNA-binding effector of Notch, and Notch is central to T-ALL, where activating NOTCH1 mutations occur in the majority of cases. A PSIP1-RBPJ row is therefore the single most promotable weak hit in this census: it connects LEDGF, the fusion's chromatin-addressing cofactor, to the dominant oncogenic pathway of a neighbouring leukaemia.
That promotability is the reason to decline it explicitly rather than leave it unwritten. An unwritten row is indistinguishable from an unexamined one, and the next beat that runs a database sweep will surface RBPJ again with nothing recorded against it.
The decline: one TAP row, in HEK293T, from an atlas, with PSIP1 as prey of a TF bait, no reciprocal, no chromatin readout, no haematopoietic cell, and no Notch experiment anywhere near it. gene-notch1 does not exist in this graph, checked against disk this cycle, so there is not even an id to attach a Notch claim to.
What would reverse it: a LEDGF co-IP recovering RBPJ from a T-ALL or KMT2A-rearranged line, or LEDGF ChIP overlapping RBPJ-bound enhancers. Neither exists.
| partner a | gene-psip1 |
|---|---|
| partner b | RBPJ |
| partner b uniprot | Q06330 |
| partner b length AA | 500 |
| partner a uniprot | O75475 |
| partner a length AA | 530 |
| verdict | DECLINED |
| adjudicated by | q160, cycle 146, 2026-09-24 |
| druggability | Not assessable. Notch pathway inhibitors (gamma-secretase inhibitors) exist and are used in T-ALL trials, but nothing here maps a PSIP1-RBPJ surface and no Notch claim is supported. |
| n methods | 1 |
| n platforms | 1 |
| n publications | 1 |
| endogenous side | no - tagged TF bait |
| intact miscore | 0.35 |
| cell state | unstimulated; the study's own split is chromatin-associated versus soluble, and IntAct does not record which fraction this row came from |
| epigenetics | None measured. The study's chromatin/soluble fractionation is the closest thing and is not resolved at row level. |
| absence class | never-measured |
| absence class note | No LEDGF co-IP in any leukaemia line has been blotted for RBPJ, and no LEDGF-RBPJ genomic overlap has been examined. gene-notch1 does not exist in this graph. |
interface residues a
[]
structures
[]
ligands
[]
records
[ "PSIP1-RBPJ, tap, association, HEK293T, PSIP1 prey, miscore 0.35, PMID 25609649" ]
sibling rows same publication
[ "MEN1-FOXB1 (tap)", "PSIP1-FOXE1 (tap)", "PSIP1-FOXK2 (tap)", "PSIP1-JUN (anti bait coip)" ]
cell systems
[ "HEK293T embryonic kidney" ]
localization
{
"compartment": "nucleus",
"subnuclear": [],
"shown_for_fusion": "NO - nobody has looked in any fusion."
}heuristic failure
{
"threshold": ">=15 distinct prey-only partners across the six baits",
"this_publication": 5,
"flagged": false,
"hand_verdict": "ATLAS - 214 purifications, 2,156 interactions",
"affected_rows_in_this_census": [
"RBPJ",
"FOXB1",
"FOXE1",
"FOXK2",
"JUN"
]
}measurements
[
{
"quantity": "count",
"subject": "prey-only partners in this census supported by PMID 25609649",
"value": 5,
"units": "partners",
"n": 212,
"universe": "the 2026-09-24 IntAct REST prey-only census across the six baits",
"null": 15,
"note": "Below the atlas threshold, so the heuristic scored these five as focused-publication support. They are not."
}
]- A Notch link would be the most citable thing in this beat and it is the least supported. Writing the decline is the only way to stop the row being rediscovered and promoted by a later sweep that finds nothing recorded against it.
- How many other publications in census.json are atlases that the >=15 threshold failed to flag? Five are confirmed for this one paper; the whole publication list has not been hand-read and the ranking column is wrong by an unknown amount.
- Did the PSIP1 recovery come from the chromatin-associated or the soluble fraction? The source resolves this and IntAct's curation does not carry it - a field worth asking IntAct to expose.
- Should this graph have a gene-notch1 entry at all? Notch is the dominant oncogenic pathway in the leukaemia immediately adjacent to this one and has no node.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
IntAct REST returns one PSIP1-RBPJ (O75475-Q06330) record: detection method 'tap', interaction type 'association', host Homo sapiens HEK293T embryonic kidney cell, PSIP1 as PREY, intact-miscore 0.35, from PMID 25609649. The same publication supplies four further prey-only rows in this census: FOXB1 and FOXE1 and FOXK2 (TAP) and JUN (anti-bait co-IP).
IntAct molecular interaction database, REST findInteractions on O75475, retrieved 2026-09-24 by q160 (partner RBPJ) · IntAct (EMBL-EBI) 2026What this evidence cannot show
One record. PSIP1 is a highly abundant, chromatin-tethered nuclear protein and is a frequent recovery in nuclear AP-MS; this corpus already records that PSIP1 is prey to nine of fifty RBP baits in an unrelated map, which is the base rate this row sits against.
Evidence that argues against this (1)
- TAP is two-step and the study reports the interactions as high-confidence after its own filtering, so this is not an unfiltered contaminant list. The objection is the atlas design and the absence of any Notch-relevant follow-up, not the purification chemistry. Li X et al. 2015
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
PMID 25609649 (Li et al., Mol Syst Biol 2015) performed tandem affinity purification followed by mass spectrometry on soluble and chromatin-associated complexes of 56 transcription factors including 37 Forkhead-box family members, carrying out 214 purifications and identifying 2,156 high-confidence protein-protein interactions, and concluded that most transcription factors form distinct protein complexes on and off chromatin.
Proteomic analyses reveal distinct chromatin-associated and soluble transcription factor complexes. · Li X, Wang W, Wang J, Malovannaya A, Xi Y, Li W, Guerra R, Hawke DH, Qin J, Chen J · Mol Syst Biol 2015What this evidence cannot show
Abstract read in full; the interaction tables were not opened. The chromatin-versus-soluble split is the paper's headline and IntAct's row does not record which fraction the PSIP1 recovery came from - a genuinely informative field that the curation loses.
Evidence that argues against this (1)
- The chromatin/soluble fractionation makes this a better-designed atlas than a whole-cell-lysate one, and if the PSIP1 recovery is from the chromatin fraction it is more interesting than this decline allows. That information exists in the source and was not retrieved, so the decline is made without it. Li X et al. 2015
the prior instance of this defect, recorded as a case and never generalised - this measures how many more there are
PSIP1 prey in an RBPJ TAP from a 214-purification TF atlas; DECLINED
the census's atlas heuristic scored this publication as focused support for five partners; it is a 2,156-interaction atlas
identical row from the same publication, graded DECLINE - the pair is written so the prior can be disagreed with