Oncofusion

start / graph / ana-q194-the-murine-coordinate-sweep-found-no-new-defects-and-one-entry-that-was-already-right

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

The murine-coordinate sweep is a NEGATIVE result - zero new defects in 3,086 entries - and its actual yield was a fourth entry that had the degron right before Q191 did

also called murine coordinate sweep · species-labelled coordinate audit · sweeping the value versus following the claim

Last cycle three entries were wrong because they used a mouse measurement as if it were a human one. A search of all 3,086 entries for the same mistake found none left. It did find a fourth entry that had already got the answer right.

The same thing, for a scientist

Q191 corrected three entries that used murine Af4 coordinates (268-290) against a human junction, and coverage item #1 owed was a sweep for the pattern. The sweep is a clean negative: **zero** residue-coordinate defects remain. It is reported as a negative rather than dressed up, because the pre-registered threshold said to. Two things came out of it that matter more than the headline. First, the detector's raw output was **8 flagged entries and 0 real defects** - a mouse-style gene symbol sitting near a number catches antibody catalogue numbers, latency in months, washout hours, genomic spans, exon ranges and table cells, so the raw count of any such sweep is uninterpretable without opening every hit. Second, sweeping the *coordinate* rather than following the *claim* found `mechanism-the-ptefb-binding-module-of-aff1-travels-on-the-reciprocal-product`, which had already placed the Siah degron on the reciprocal product and used the 347 junction - i.e. it reached Q191's conclusion first, and Q191 missed it and reported that only one entry had been right.

Go deeper
The result, stated as the negative it is

Pre-registered before looking: >=10 undeclared coordinates means systemic and needs a schema rule; 1-9 means fix individually; 0 means the Q191 cluster was the whole population. The adjudicated count is 0, so the third branch applies.

Two independent measures agree:

  • Forward scan. 40 of 3,086 entries carry any non-human-coordinate signal; 15 are CLASS A (a Title-case mouse symbol next to a number); 8 were undeclared; 0 survive adjudication as residue coordinates whose own species is non-human.
  • Recall bound. Every entry in the corpus carrying a known-murine coordinate - 268-290 (Af4 Siah element, 5 entries) and 277-299 (Fmr2 element, 1 entry) - declares its murine origin. 0 bare.

The sweep does not prove the corpus is free of the error; it proves the error is not present in any entry reachable by either route. A murine number written under a human symbol with no species word anywhere would still be invisible, which is why the recall bound was run separately against the specific values rather than left to the pattern detector.

8 flags, 0 defects: what a coordinate detector actually catches

The naive version of this sweep is worthless and it is worth recording why, because the obvious implementation looks fine. All 3,086 entries match a bare \d{2,4}-\d{2,4} (publication years, journal page ranges, PMIDs, genomic spans) and 1,569 contain a species word somewhere in the file, because mouse models are legitimately everywhere in this corpus. Requiring explicit residue language plus mouse-symbol casing cut that to 8. All 8 were still false:

flagged textwhat it actually is
Mll C-terminus (Millipore 05-765an antibody catalogue number
Men1 ablation | 2-6tumour latency in months
Meis1 does not rebound within 72-96washout hours
Meis1** chr11:18,875,428-19a genomic coordinate
Mll exons 5-11exon numbers, not residues
Meis1*** |\n| 18-20a table cell
aa 544-553a HUMAN coordinate (see below)

So the adjudication had to be encoded, not asserted: a defect requires the species word within 40 characters of the coordinate and no human reference within 80. That is the only reason the final 0 is reproducible rather than my opinion.

The one candidate that survived, and why it was still wrong

mechanism-nes-crm1-chromatin-addressing carried CALM NES aa 544-553 in a comparison table whose adjacent cell read "murine bone marrow; a CRM1-AF10 chimera is leukaemogenic". The species word was 20 characters away and belonged to the next column.

The owning entry, mechanism-calm-af10-crm1-nuclear-export-requirement, settles it: it says "CALM residues 544-553 form a leucine-rich NES recognised by CRM1/XPO1" and carries Q13492 - human PICALM. The coordinate is human; the assay was murine. A detector cannot tell the species of a coordinate from the species of the experiment it appears next to, and in a table they sit centimetres apart. Fixed by adding the accession to the cross-reference row, which is the whole defect.

The actual yield: following a claim finds citations, sweeping a value finds entries

Q191 found the degron error by following one claim through the entries that cite each other, and concluded that one entry - fusion-kmt2a-aff1 - had been right. That was incomplete.

Grepping the coordinate 268-290 across all 3,086 entries returned five, one of which Q191 had never looked at: mechanism-the-ptefb-binding-module-of-aff1-travels-on-the-reciprocal-product. It uses the 347 junction throughout and lists under reciprocal_product_carries: "AFF1 1-346, NHD + ALF, cyclin T1-binding sequence 1-308, Siah-binding degron 268-290". It had already put the degron on the reciprocal product and off the forward fusion - Q191's conclusion, reached earlier and by a different route. It even hedged the number: "268-290 (as recorded elsewhere in this graph)".

So the corpus held the right answer in three places (this entry, fusion-kmt2a-aff1, and protein-aff1-domains' ENST00000307808 derivation) while three others held the wrong one. The error was never corpus-wide; it was localised to the subtraction-* cluster and the beat that inherited from it. Q191's framing of 'three entries said it keeps it' was right about the wrong entries and wrong about how isolated they were.

The transferable method point: following a claim traverses the citation graph, so it finds the entries that already talk to each other - which are disproportionately the ones that share the error. Sweeping the literal value finds the entries that never joined that conversation, and those are where an independent answer lives.

Facts and figures
entries scanned3086
entries with any signal40
class a hits15
raw undeclared flags8
adjudicated defects0
known murine coords undeclared0
detector precision on raw flags0 of 8
preregistered thresholds>=10 systemic; 1-9 isolated; 0 negative result
verdictNEGATIVE RESULT
spike controlpre-correction subtraction-aff1-control and subtraction-synthesis-how-general from commit abc6fe4 both FLAGGED; the detector catches a verified known-positive
false positive kinds
[
  "antibody catalogue number",
  "tumour latency in months",
  "washout hours",
  "genomic coordinate",
  "exon range",
  "table cell",
  "human coordinate beside a murine assay"
]
corrections made
[
  "mechanism-nes-crm1-chromatin-addressing (accession added)",
  "mechanism-the-ptefb-binding-module-of-aff1-travels-on-the-reciprocal-product (species labelled)",
  "ana-q191-... (fourth correct entry credited)"
]
entries holding the correct degron answer before q191
[
  "fusion-kmt2a-aff1",
  "mechanism-the-ptefb-binding-module-of-aff1-travels-on-the-reciprocal-product",
  "protein-aff1-domains"
]
Still open
  1. Would this detector find a coordinate defect in a data field with no surrounding prose? The 40- and 80-character windows assume prose context, and much of this corpus's coordinate content lives in data objects.
  2. Should the schema require a species and a reference accession on every residue coordinate? The sweep says the error is not currently present, which is an argument for a cheap rule rather than against one - the Q191 cluster shows what it costs when it does happen.
  3. How many other corpus-wide claims were settled by following a citation chain rather than sweeping a value? Q191 is one confirmed case where the two methods gave different answers.
  4. Does mechanism-calm-af10-crm1-nuclear-export-requirement state CALM 544-553 in the numbering of the primary paper? The accession makes it human; nobody has checked it against the source.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.internal

    A sweep of all 3,086 entries for non-human residue coordinates yields 40 entries with any signal, 15 CLASS A hits, 8 raw undeclared flags and 0 adjudicated residue-coordinate defects; separately, all 6 entries carrying a known-murine coordinate (268-290, 277-299) declare its murine origin.

    What this evidence cannot show

    The detector keys on Title-case mouse symbols from a fixed token list plus explicit residue language. A murine coordinate written under a human symbol with no species word anywhere would be missed, which is why the recall bound was run against the specific known values. The token list is not exhaustive and the recall bound covers only the two murine coordinates this project is known to carry.

    Evidence that argues against this (1)
    • A negative result from an instrument with 0/8 precision on its own raw output should not be over-trusted in the other direction either. The same looseness that produced 8 false positives means the token list and the 40/80-character windows were tuned against one known case - the Q191 cluster - and a differently-shaped instance (a coordinate in a data field with no prose around it, or a rat or zebrafish number under a symbol not in the token list) is exactly what this design would not see. Oncofusion KG 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The single candidate surviving the encoded adjudication, CALM NES aa 544-553, is a HUMAN coordinate: the owning entry states CALM residues 544-553 form a leucine-rich CRM1/XPO1-recognised NES and carries UniProt Q13492 (human PICALM), while the adjacent table cell's 'murine bone marrow' describes the transformation assay.

    What this evidence cannot show

    Resolved from this corpus's own owning entry plus the accession it carries; the primary paper's numbering was not re-opened this cycle. If that entry's 544-553 is itself mis-transcribed, this adjudication inherits the error - the check confirms the coordinate is human, not that it is correct.

    Evidence that argues against this (1)
    • The owning entry never uses the word 'human' - it contains only 'murine' and 'mouse' - and the human reading rests entirely on the presence of the Q13492 accession. An accession in a citation block is weaker evidence of the numbering used in the prose than an explicit statement would be, and by the standard this sweep is enforcing, that entry should say so outright. UniProtKB 2026
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.internal

    mechanism-the-ptefb-binding-module-of-aff1-travels-on-the-reciprocal-product uses the AFF1 residue-347 junction and lists the Siah-binding degron 268-290 among what the RECIPROCAL product carries, so it had reached Q191's conclusion - degron off the forward fusion - before Q191 was written, and Q191 did not cite it.

    What this evidence cannot show

    Corpus-internal. Found by grepping the coordinate, not by re-deriving the biology. That entry's status is 'verified' with confidence 'moderate', and it quoted the murine 268-290 for a human product without conversion - a labelling imprecision that does not change its conclusion, because the reciprocal carries AFF1 1-346 which contains the element on either numbering. Now labelled.

    Evidence that argues against this (1)
    • Crediting this entry as 'already right' flatters it slightly. It placed the degron on the reciprocal as part of a domain-partition table, without arguing the retained-versus-lost question or noticing that other entries in the corpus asserted the opposite. Holding the correct value in a table is not the same as resolving a dispute, and the dispute is what Q191 actually settled. Oncofusion KG 2026
Connected to