Re-grade the interactome on PSI-MI method class instead of bait role - a dry experiment, and an EXECUTION of a prescription data/symbol-collisions.json already wrote down in cycle 57 and nobody has run
also called method-class re-grade · bait-free partner read
Sort the protein partners by what kind of experiment found them rather than by which protein carried the tag, then read the two dozen the old sorting threw away, and see whether the order changes.
The same thing, for a scientist
**Design.** Two arms, both dry, both on data already pulled into `analyses/q160-the-prey-only-interactome/rest-roles.json` (879 IntAct REST records, six baits). **Arm A, re-grade:** assign every record a method class from its PSI-MI `detectionMethod` - `biophysical-direct` (ITC, NMR, x-ray diffraction, molecular sieving, surface plasmon resonance), `ap-ms` (anti-bait/anti-tag co-IP, TAP), `proximity` (BioID, APEX), `display` (phage display, peptide array), `genetic` (two-hybrid and its variants), `crosslink`. Re-rank the 212 human prey-only partners on method class and count how far the top 20 moves against the current role-based ranking. **Arm B, read:** fetch and read the source paper for each of the 23 unread never-prey partners, and score whether its title or abstract names one of the six baits. **Pre-registered decision rule: if fewer than 4 of the 24 bait-free partners have a source paper naming one of the six baits, `hyp-q160-...` is dead and this entry says so.** [E1]
PRIOR ART: this design was already prescribed and never executed
This is not a new idea and the entry should not read as one. data/symbol-collisions.json, row 'IntAct curates enzyme-substrate assays as interactions, so a partner census counts substrates as partners', added 2026-09-22 by king-cycle-57 from Q104, closes with:
Prescription: Filter the partner census by DETECTION METHOD, excluding enzymatic-assay and structural-fragment evidence, rather than by symbol class. Record the denominator effect when announcing the change: it is small.
That is Arm A. I arrived at it from a different direction - the bait/prey role column rather than the substrate problem - and the registry got there first by two cycles 2.
Three things the prior art adds that this design did not have:
- The denominator effect is already measured and it is small: excluding the histone and ribosomal class moves coverage 43.8% to 46.1%. So a method-class filter should be justified on the object, not on the arithmetic, and Arm A must report the coverage change rather than lead with it.
block_autogeneration: Trueis set on that registry row, so any automated pass over the partner census is supposed to stop at it.- The row names the cleanest cases to score first - H3C13 entering as a DOT1L partner by methyltransferase assay, and the 11 KMT2A-H3C1 ITC/NMR/X-ray rows on a cassette every fusion deletes.
What this entry adds to the prescription: a second axis (the bait-free role column), a pre-registered decision rule, a matched control, and a third pre-registered outcome. The prescription said what to filter on; it did not say how to score whether the filter was an improvement.
Why it is worth doing and why it is cheap
Cost: hours, no reagents, no external data pull. The records are on disk with detectionMethod already captured; the 23 papers are 23 PubMed fetches.
What it settles that nothing else does. The project has re-measured interactome coverage five times and has never re-measured the grading that coverage is computed over. Rule 90's lesson was that a number which closes a line of work carries the highest duty of reproducibility; the bait/prey grade is that number's denominator and has never been audited.
Pre-registered, because this beat has a stake in the answer. The hypothesis is this beat's own, the two supporting examples are this beat's own entries, and the temptation to score ambiguous papers generously is obvious. The decision rule and the scoring criterion - does the title or abstract name one of the six baits - are fixed here before the reading, and the criterion is deliberately crude so that it cannot be bent case by case.
A third outcome is pre-registered too, because neither 'the grade is fine' nor 'the grade is broken' may be right. The likely result is that the bait-free class contains both AKAP8-shaped focused biophysics and BRET-screen noise, in which case the correct conclusion is that method class and bait role are orthogonal and triage needs both - not that either one wins.
| design | Arm A: map every PSI-MI detectionMethod in rest-roles.json to one of six method classes, re-rank the 212 human prey-only partners, and report rank displacement of the current top 20 plus the full mapping table. Arm B: fetch and read the source publication for each of the 23 unread never-prey prey-only partners and score, by the pre-registered crude criterion, whether its title or abstract names KMT2A, AFF1, AFF4, MEN1, PSIP1 or DOT1L. |
|---|---|
| readout | Arm A: Spearman correlation and maximum rank displacement between the role-based and method-based rankings of the 212. Arm B: k of 24 bait-free partners whose source paper names one of the six baits in title or abstract. |
| controls | Arm A control - apply the same re-grade to the 71 bait-SUPPORTED partners; if their ranking moves as much as the prey-only one, the method class is reordering everything and says nothing specific about the discard bin. Arm B control - score 24 randomly drawn partners from the 188 that DO have a literal prey role, by the identical criterion, as the base rate. Without that base rate, Arm B's k is the same kind of uninterpretable proportion rule 84 records. |
| cost tier | low |
| timeline | hours; no new data pull, no reagents |
| prereg decision rule | If fewer than 4 of the 24 bait-free partners have a source paper naming one of the six baits in title or abstract, hyp-q160-the-bait-prey-grade-discards-the-bait-free-biophysics-that-is-the-best-evidence-in-the-tail is REFUTED and must be marked so. If the bait-free k does not exceed the k of the 24 matched prey-role controls, the hypothesis is likewise refuted, because the class is then not enriched. |
| prereg third outcome | NEITHER. If the bait-free class contains both focused biophysics and screen noise at similar rates to the prey-role class, the conclusion is that method class and bait role are orthogonal and triage should carry both columns - not that either grading wins. |
| prior art | data/symbol-collisions.json row 'IntAct curates enzyme-substrate assays as interactions...', added 2026-09-22, prescribes Arm A. This entry executes and extends it; it does not originate it. |
measurements
[
{
"quantity": "count",
"subject": "IntAct REST records available on disk for the re-grade",
"value": 879,
"units": "records",
"n": 879,
"universe": "rest-roles.json, six baits plus two KMT2A taspase1 chains, de-duplicated by interaction AC, 2026-09-24",
"null": null,
"note": "KMT2A 315, AFF4 171, MEN1 136, PSIP1 114, AFF1 81, DOT1L 62."
}
]- Both arms are scored by the beat that proposed the hypothesis. The pre-registered decision rule and the matched control are the only protection against that, and a later reader should check that the control was actually run rather than trusting that it was specified.
- Is there a standard PSI-MI method-class hierarchy that should be used instead of this beat's six-class mapping? The PSI-MI ontology has a tree structure under MI:0001 that was not consulted.
- Should the re-grade run over PSICQUIC as well, so that the endpoint effect and the grading effect can be separated? They are currently confounded in every comparison this beat makes.
- The registry prescribes excluding structural-fragment evidence; this beat's hypothesis argues that the ITC and NMR biophysics is the tail's BEST evidence. Those are opposite instructions on the same records. Which is right depends on whether the fragment measured is in a region the fusion retains - the H3C1 PHD3-bromodomain case says sometimes it is not.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
analyses/q160-the-prey-only-interactome/rest-roles.json holds 879 de-duplicated IntAct REST interaction records across the six core baits plus the two KMT2A taspase1 processed chains, each carrying detectionMethod, interaction type, experimentalRole for our protein, host organism, publication PMID and intact-miscore. 24 of the 212 human prey-only partners have no record in which our protein's role is 'prey'; one of those 24 (AKAP8) has been read in full by this beat and 23 have not. H3C1 was also read but is not one of the 24: it carries eleven bait-free records and two prey records.
analyses/q160-the-prey-only-interactome/census.json and rest-roles.json, generated 2026-09-24 · this project 2026What this evidence cannot show
The record set is IntAct only. PSI-MI detection-method terms are a controlled vocabulary but their mapping to the six method classes proposed here is this beat's own and is not standard; a different mapping could give a different re-ranking, which is why Arm A's output must include the mapping table.
Evidence that argues against this (1)
- A re-grade on method class has its own failure mode: it will promote in vitro peptide-display and ITC rows that measure a fragment rather than a protein, which is exactly the objection this beat raised against the SDCBP row. A method-class ranking is not obviously better than a role ranking, only different, and Arm A can only show that they differ. this project 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
data/symbol-collisions.json carries the row 'IntAct curates enzyme-substrate assays as interactions, so a partner census counts substrates as partners', added 2026-09-22 by king-cycle-57 from Q104, with block_autogeneration set to true. Its prescription is: filter the partner census by DETECTION METHOD, excluding enzymatic-assay and structural-fragment evidence, rather than by symbol class, and record the denominator effect when announcing the change because it is small - excluding the histone and ribosomal class moves coverage from 43.8% to 46.1%.
What this evidence cannot show
Project-internal registry. The 43.8% to 46.1% figures are that row's own measurement over a 122-partner missing set and are NOT this beat's 212-partner prey-only denominator, so they are not directly comparable and Arm A must recompute the effect on its own population.
Evidence that argues against this (1)
- The registry prescribes excluding enzymatic-assay and structural-fragment evidence, which would REMOVE the ITC/NMR biophysics that this beat's hypothesis argues is the tail's best evidence. The two prescriptions point in opposite directions on the same records, and Arm A has to resolve that rather than assume they agree. this project 2026
two dry arms with a pre-registered decision rule and a matched control
runs entirely on rest-roles.json, already on disk