Oncofusion

start / graph / ana-gothe-2019-full-text-the-route-that-worked

Analysisestablished confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.Q107-break-machinery-deposition

Gothe 2019's full text IS reachable - not through PMC, which has no deposit, but through an author search that surfaces the bioRxiv preprint; fourteen routes, and the two Europe PMC id links that look like a PMC record are both traps

also called the Gothe retrieval audit · pubmed_pmc_refs is not pubmed_pmc · Q107 route audit

A previous cycle concluded this paper's full text could not be reached and gave up. It can: search the first author's name in Europe PMC and the free preprint version appears, with the complete methods. Two of the obvious shortcuts return long lists of identifiers that look like the paper's own record and are actually the papers that cite it.

The same thing, for a scientist

PMID 31202576 has **no PMC deposit and no Europe PMC full text**: the core record returns `pmcid: None`, `isOpenAccess: N`, `inEPMC: N`, `inPMC: N`, `hasPDF: N`, `hasSuppl: N`, `fullTextIdList: None` and a single `fullTextUrl` marked *Subscription required*; `PMC/31202576/fullTextXML` is a 404; and `BODY:"Spatial Chromosome Folding and Active Transcription"` returns **hitCount 0** [E1]. **Two routes return large, plausible-looking id lists and neither is the paper.** `elink dbfrom=pubmed db=pmc id=31202576` returns **103 ids under linkname `pubmed_pmc_refs`** and **no `pubmed_pmc` linkset at all** - `pubmed_pmc_refs` is *articles in PMC that cite this one*. And `esearch db=pmc` on the exact title phrase returns **count 98** with a clean non-empty querytranslation and `warninglist: None` - 98 citing papers whose *reference lists* contain the title string [E1]. **The route that worked is `AUTH:"Gothe HJ"` in Europe PMC: hitCount 8, and row six is `PPR63701`, the December 2018 bioRxiv preprint of the same title**, DOI 10.1101/485763, CC-BY-NC-ND, whose `published_doi` the bioRxiv API confirms as 10.1016/j.molcel.2019.05.015 [E2]. Its full text, including the complete STAR-Methods-equivalent, is 179,145 characters and was retrieved and grepped locally [E2].

Go deeper
Every route tried, and what each returned
#routeresult
1Europe PMC search DOI:"10.1016/j.molcel.2019.05.015" resultType=corehitCount 1. pmcid: None, isOpenAccess: N, inEPMC: N, inPMC: N, hasPDF: N, hasSuppl: N, fullTextIdList: None, citedByCount: 142. Sole fullTextUrl: DOI, availability "Subscription required"
2Europe PMC PMC/31202576/fullTextXMLHTTP 404
3Europe PMC BODY:"Spatial Chromosome Folding and Active Transcription"hitCount 0
4esearch db=pmc on the exact title phrasecount 98, querytranslation "Spatial Chromosome Folding and Active Transcription Drive DNA Fragility"[All Fields], warninglist: None. All 98 are citing papers - PMC full-text search matches reference lists
5elink dbfrom=pubmed db=pmc id=31202576103 ids, linkname pubmed_pmc_refs only. No pubmed_pmc linkset exists, i.e. no PMC record
6Cell Press, URL I constructed as .../molecular-cell/fulltext/S1097-2765(19)30364-2HTTP 403. The PII was invented and is wrong - see "my own errors" below
7Unpaywall on the published DOIis_oa: true, oa_status: **bronze**, one location: publisher, publishedVersion, http://www.cell.com/article/**S1097276519303879**/pdf
8curl that publisher PDF with a browser UAHTTP 403, 5,725 bytes of Cloudflare "Just a moment" interstitial. Unpaywall says OA; the link is not machine-retrievable
9Semantic Scholar graph API by DOI{"error":"Paper with id DOI:10.1016/j.molcel.2019.05.015 not found"}
10Europe PMC AUTH:"Gothe HJ"hitCount 8. Row six = PPR63701, source PPR, 2018, the preprint of the same title. THE ROUTE THAT WORKED
11Europe PMC core on EXT_ID:PPR63701 AND SRC:PPRDOI 10.1101/485763, 20 authors, corresponding author Vassilis Roukos (IMB Mainz), inEPMC: N, hasPDF: N, Unpaywall OA PDF link
12bioRxiv API get_preprint on 10.1101/485763success; published_doi: 10.1016/j.molcel.2019.05.015, license cc_by_nc_nd, version 1, 2018-12-03, category cell biology, jatsxml URL
13curl the bioRxiv PDF, the .source.xml JATS, and connect.biorxiv.org/biorxiv_xml/485763HTTP 429 error code: 1015 (Cloudflare rate limit) on the first two, repeatedly and with several user agents; 404 on the third
14WebFetch, then a text-extraction proxy, on biorxiv.org/content/10.1101/485763v1.fullfirst 429 with Retry-After: 39, then HTTP 200. The proxy returned 179,145 characters of markdown including the methods

bioRxiv blocks direct programmatic fetches from this environment with Cloudflare 1015 but serves the same URL to a browser-shaped client. That is the single fact that turns "unreachable" into "reachable", and it is worth writing down because it will apply to every bioRxiv preprint this project ever needs.

Scope of the negative that remains true: the published July 2019 Mol Cell version is not retrievable from here by any of these fourteen routes, and there is no PMC or Europe PMC full text of it. Everything read is the December 2018 preprint, version 1. That bound is stated on every claim drawn from it.

Two identifier traps, both of which return a well-formed wrong answer

Trap 1 - pubmed_pmc_refs is not pubmed_pmc. elink dbfrom=pubmed db=pmc returned a linksetdbs array containing exactly one entry, linkname: "pubmed_pmc_refs", with 103 ids beginning 13503036, 13381955, 13060179, .... A consumer reading linksets[0].linksetdbs[0].links[0] gets 13503036 and would report it as Gothe's PMC id. It is some other paper that cites Gothe. The presence of a db=pmc linkset does not mean the article is in PMC; you must check the linkname. In this project's own corpus, PMC ids for the other papers in this dispute were obtained from esummary db=pubmed articleids (which correctly returns pmc: PMC8170508 for Canela 2019, PMC6133249 for Canela 2017, PMC10591855 for Gray 2023, PMC8511143 for Atkin 2021, PMC5550487 for Gomez-Herreros 2017 - and no pmc id at all for Gothe 2019), which is the safe route 1.

Trap 2 - a PMC full-text search matches reference lists. esearch db=pmc on a 12-word exact title phrase returned 98 records. The querytranslation was non-empty and the warninglist was None, so by this project's own rule-4 test the instrument was working perfectly. The 98 are citing papers. A PMC hit count on a title string is a citation count, not a deposit check. The deposit check is the pmc field of esummary db=pubmed, or inPMC from Europe PMC core.

And a third, smaller one. Unpaywall reported is_oa: true with oa_status: bronze for the published DOI. Bronze means free-to-read on the publisher's site without a licence, and in this case the publisher serves it behind a Cloudflare challenge. is_oa: true is therefore not a statement that a machine can fetch it.

My own errors in this run, both of the same shape

1. I constructed a Cell Press PII instead of fetching one. I wrote https://www.cell.com/molecular-cell/fulltext/S1097-2765(19)30364-2 from the journal's usual pattern and the article's page range. It 403'd, and I initially read that as a paywall. The real PII, which Unpaywall supplied two calls later, is S1097276519303879. So my 403 was ambiguous between "paywalled" and "does not exist", and I could not have told which. A publisher URL must come from a resolver - Unpaywall, Crossref, the DOI redirect - never from the journal's URL grammar. This is the same defect as inventing a PMID.

2. I overwrote a downloaded file inside a loop and then grepped the wrong one. A for u in ...; do curl -o $SP/t.txt "$u"; done loop fetched the full text successfully and then overwrote t.txt with a 17-byte Cloudflare error from the next URL. My first grep pass over the copied file reported 0 occurrences of every single token - CTCF, cohesin, TK6, K562, Methods, all zero. Caught because a corpus of that size returning zero for Methods is impossible. A zero across every token in a token list is an instrument failure, not a finding, and it is the file-level twin of this project's empty-querytranslation rule.

3. The one that would have gone into the entry. The summarising fetch of the preprint reported, as its answer to the CTCF/cohesin question: "No depletion/knockdown/knockout experiments found." On the raw text, degron appears 5 times, auxin 13, siRNA 10, shRNA 12, depletion 5 and knockdown 2. The summariser's conclusion happened to be right about CTCF and cohesin specifically and was wrong as stated. A summariser's negative about a methods section is not evidence; re-derive it by token count on the raw text and read every hit in context. That is what turned this beat's Gothe verdict from a paraphrase into a bounded claim - and it is the difference between the finding and the thing this beat was explicitly told not to do.

Facts and figures
scriptfourteen retrieval routes executed by hand; no committed script
statusrun
result summaryNo PMC deposit and no Europe PMC full text for PMID 31202576; the published Cell Press PDF is Unpaywall-'bronze' but Cloudflare-blocked. The full text IS reachable as the bioRxiv preprint 10.1101/485763, found via Europe PMC AUTH:"Gothe HJ" (8 hits, row 6), and retrieved as 179,145 characters including methods by a browser-shaped fetch after direct curl was rate-limited with Cloudflare error 1015. Two identifier traps recorded: elink's pubmed_pmc_refs linkname returns citing papers and looks like a PMC record, and esearch db=pmc on a title phrase returns 98 citing papers through their reference lists with a clean querytranslation and no warninglist.
verdictreachable-via-preprint; published-version-not-reachable-unauthenticated
inputs
[
  "NCBI esummary db=pubmed, elink db=pmc, esearch db=pmc",
  "Europe PMC REST search (DOI, BODY, AUTH, EXT_ID), fullTextXML endpoint",
  "Unpaywall API v2 on 10.1016/j.molcel.2019.05.015 and 10.1101/485763",
  "Semantic Scholar graph API",
  "bioRxiv API get_preprint",
  "biorxiv.org/content/10.1101/485763v1.full"
]
How this could bite us back
  • Building a bounded absence on a summariser's reading of a methods section would have produced a claim that is right by accident. The token-count re-derivation is the only reason this beat's Gothe negative is safe to quote, and it should be a precondition for every future methods-level negative in this graph.
Still open
  1. Does the published July 2019 version differ from the December 2018 preprint in any way that affects the CTCF/cohesin negative? One authenticated library fetch of sciencedirect.com/science/article/pii/S1097276519303879 would settle it.
  2. How many other papers this graph cites from its abstract alone have a reachable preprint under AUTH:"<first author>"? This project has repeatedly recorded 'no PMC deposit' as the end of the road; the AUTH: route was never tried, and it took one call.
  3. Is bioRxiv's Cloudflare 1015 response stable, or was it transient rate limiting from repeated attempts in one minute? If it is per-IP rate limiting rather than a block, a slower single fetch may work with plain curl and the proxy would be unnecessary.
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    PMID 31202576 has no PMC record: esummary db=pubmed returns articleids containing only pubmed and doi with no pmc entry (while returning PMC8170508 for 31202577, PMC6133249 for 28735753, PMC10591855 for 37788669, PMC8511143 for 34405474 and PMC5550487 for 28794467), Europe PMC resultType=core returns pmcid None with inEPMC N, inPMC N, isOpenAccess N, hasPDF N, hasSuppl N, fullTextIdList None and a sole fullTextUrl marked 'Subscription required', PMC/31202576/fullTextXML returns HTTP 404, and BODY:"Spatial Chromosome Folding and Active Transcription" returns hitCount 0; meanwhile elink dbfrom=pubmed db=pmc returns 103 ids under linkname pubmed_pmc_refs with no pubmed_pmc linkset, and esearch db=pmc on the exact title phrase returns count 98 with a non-empty querytranslation and warninglist None, all 98 being citing papers matched through their reference lists.

    What this evidence cannot show

    Absence of a PMC record is a fact about 2026-09-22; Cell Press deposits some NIH-funded author manuscripts on a delay, and this paper's funding is European (ERC/IMB Mainz), which is consistent with no PMC deposit ever appearing. The BODY: zero is bounded by Europe PMC's full-text index, which only covers articles it holds, so it could not have found this paper whether or not the phrase is in it - it is a consistency check on route 1, not independent evidence. The 98-record PMC count was not screened article by article; the inference that all 98 are citing papers rests on the 103-id pubmed_pmc_refs linkset and on the absence of a pubmed_pmc linkset, plus spot agreement between the two id lists.

    Evidence that argues against this (1)
    • The one route not tried is an institutional or library proxy, which would very likely serve the published version and its STAR Methods directly. So 'unreachable' here means unreachable from an unauthenticated machine, not unreachable to this lab - an Iqbal Lab member with Northeastern credentials can almost certainly open sciencedirect.com/science/article/pii/S1097276519303879 and settle the preprint-versus-published question in a minute. That is the cheapest open action arising from this entry. Unpaywall (OurResearch) 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.preprint

    Europe PMC AUTH:"Gothe HJ" returns hitCount 8 including PPR63701, source PPR, 2018, titled 'Spatial chromosome folding and active transcription drive DNA fragility and formation of oncogenic MLL translocations' with DOI 10.1101/485763, twenty authors ending in Roukos V, corresponding author Vassilis Roukos at the Institute of Molecular Biology (IMB) Mainz, licence cc_by_nc_nd, version 1, dated 2018-12-03; the bioRxiv API confirms its published_doi as 10.1016/j.molcel.2019.05.015; direct curl of its PDF and of its .source.xml JATS return HTTP 429 'error code: 1015' from Cloudflare, while a browser-shaped fetch of biorxiv.org/content/10.1101/485763v1.full returns HTTP 200 and 179,145 characters of full text including methods.

    DOI
    What this evidence cannot show

    Version 1 is the only version bioRxiv reports, and it predates the Mol Cell publication by seven months. The retrieved text is a markdown rendering by a third-party extraction proxy, not the publisher's XML, so figure images are absent, table formatting is lost, and supplementary files were not retrieved; all claims drawn from it are from body text, figure legends and the methods, which are present in full. The 179,145-character count includes reference anchor text and site boilerplate, so any token count over it is an upper bound on body-text occurrences.

    Evidence that argues against this (1)
    • A preprint is not the paper, and this project has a standing rule against presenting one as the other. The specific risk for the CTCF/cohesin negative is a revision-added experiment, and the seven-month gap is real - the CD34+ sBLISS samples were deposited in March 2019, after this preprint. What limits that risk is the GEO deposit, finalised in June 2019 at publication, which contains no CTCF or cohesin sample either: an occupancy-perturbation sequencing experiment added in revision would have left a GEO footprint and did not. NCBI Gene Expression Omnibus 2019
Connected to