ArrayExpress experiment → study¶
Needs: an E-MTAB-style accession Feeds: study → runs → files
Routes: 1 measured, with a fallback that is not yet a declared route.
Route¶
ae_experiment->study:idf_secondary reads Comment[SecondaryAccession] from the study's
IDF file at https://www.ebi.ac.uk/biostudies/files/<acc>/<acc>.idf.txt. One request.
Measured¶
Corpus: arrayexpress-ena — every ArrayExpress study BioStudies records an
ENA link for, enumerated on 2026-09-15 and committed. Surveyed: 2026-09-15.
The population is the point. BioStudies independently says an ENA link exists for each of these, so a study whose IDF declares no secondary accession is a countable recall failure rather than an unknown.
| Count | ||
|---|---|---|
| Resolved | 20,137 | 97.3% |
| No secondary accession declared | 551 | 2.7% |
| Failed | 5 | 0.02% |
| Distinct studies found | 20,138 |
It is not one route, it is two populations¶
| Prefix | What it is | Total | Resolved | Rate |
|---|---|---|---|---|
E-MTAB |
native ArrayExpress submissions | 9,802 | 9,777 | 99.7% |
E-GEOD |
ArrayExpress's imports of GEO series | 10,597 | 10,086 | 95.2% |
E-ERAD |
the Sanger ERAD collection | 274 | 193 | 70.4% |
| others | SYBR, GEUV, TABM | 20 | 20 | 100% |
For a study submitted to ArrayExpress the IDF route is essentially exact. For
a GEO series mirrored into ArrayExpress it is not, and the right answer for
those is to go to GEO — where the SOFT family file resolves 99.8%
— rather than to read a derived record of a derived record. E-ERAD is the
weakest at 70.4% and the smallest, at 274 studies.
This replaces a 2026-03 figure of 7,179 queried and 457 empty, whose population was never recorded and so cannot be compared with anything.
Five studies exist without existing¶
The 5 failures are all E-MTAB, and all are genuine 404s on the IDF. The study
records themselves are live — /api/v1/studies/E-MTAB-14460 returns a title and
a ReleaseDate of 2026-01-26 — and the search index returns them. They simply
register no files at all: no IDF, no SDRF, which are the two files an
ArrayExpress study is defined by. Eight months past their stated release date.
E-MTAB-14460 E-MTAB-15098 E-MTAB-15118 E-MTAB-15165 E-MTAB-15306
The fallback¶
E-MTAB-6505 declares no Comment[SecondaryAccession], but its SDRF still carries
Comment[BioSD_SAMPLE] BioSample accessions, and querying ENA for one of them
(SAMEA5053920) recovers ERR2861957 / PRJEB29431 / ERP111731.
So the chain is: IDF secondary accession → else SDRF BioSamples → ENA. Both hops
are declared routes (ae_experiment->biosample:sdrf, biosample->run:ena_filereport)
and fetch_series.relations walks the composition for any ArrayExpress entry
point.
How much the fallback actually recovers: 4.2%¶
Measured over exactly the 551, on 2026-09-15
(corpus arrayexpress-no-secondary, committed):
| Prefix | Studies | SDRF names a BioSample | |
|---|---|---|---|
E-GEOD |
450 | 0 | 0% |
E-ERAD |
81 | 16 | 20% |
E-MTAB |
20 | 7 | 35% |
| total | 551 | 23 | 4.2% |
Not one of the 450 E-GEOD studies carries a Comment[BioSD_SAMPLE]. That
is not a near miss, it is a categorical property: an ArrayExpress import of a
GEO series has a GEO-derived SDRF, and GEO-derived SDRFs do not carry BioSample
columns. 15 of the 450 could not be read at all.
The fallback is therefore real but narrow — and complete where it applies. The 23 studies yield 756 BioSamples, and the second hop resolves every one of them: 756 of 756 to 1,547 runs, nothing empty, nothing failed. So it rescues E-MTAB-6505 and 22 other studies outright, and it is simply the wrong tool for the group that dominates the failures.
The right route for an E-GEOD study is GEO. E-GEOD-63923 is GSE63923;
the GEO series resolves through the SOFT family file
at 99.8%. Reading a derived record of a derived record, and then failing over to
a second derived field of the same derived record, is two ways of avoiding the
archive that actually holds the answer.
A trap in the file list¶
The BioStudies file API pages silently. With no explicit limit it returns the first 25 of
however many files a study registers, which for E-MTAB-9221 would have declared 15 of its
40 registered FASTQs unregistered. Always pass a limit; biostudies.registered_files pages
at 500.
That matters because whether a study registers a file decides whether its declared URI
can be trusted — see the E-MTAB-8060 versus E-MTAB-9221 case in
fetch10xmeta,
which is the milestone-2 problem.