ArrayExpress experiment to file

Route id: ae_experiment->file:sdrf Provider: EBI BioStudies Direction: E-MTAB/E-GEOD experiment → submitter fastq URLs, keyed by run Status: implemented; tier one passed, census pending

The only route to the submitter's own fastq files for an ArrayExpress study. ENA's fastq_ftp holds ENA's derivations and submitted_ftp holds whatever format was deposited; neither is the submitter's fastq as named in the study.

Requests

Two, and the second is not optional:

curl -sL 'https://www.ebi.ac.uk/biostudies/files/E-MTAB-9221/E-MTAB-9221.sdrf.txt'
curl -sL 'https://www.ebi.ac.uk/biostudies/api/v1/studies/E-MTAB-9221/files?limit=500&offset=0'

The SDRF supplies the URIs. The file listing decides which of them to believe — see SDRF URIs point at a decommissioned mirror, which is the reason this route costs two requests instead of one.

Parsing

URIs are found by shape, not by column name. SDRF column headers are not standardised across studies, and the ones that matter repeat: E-MTAB-9221 has two Comment[FASTQ_URI] columns, one per mate. A row is attributed to whatever run accession it mentions.

Tier one

Study	Result	Why
E-MTAB-9221	40 files	registers all 40; URIs rewritten to BioStudies
E-MTAB-6505	8 files	registers 8 fastqs, all kept
E-MTAB-5448	36 files	URIs are already on ftp.sra.ebi.ac.uk, so untouched
E-MTAB-8060	empty	registers no fastq; every mirror URI dropped, run falls through to the ENA BAM
E-MTAB-9216	empty	declares no fastq URIs at all

Five verdicts, no failures, and the two empties are the correct answer rather than a gap.

Measured

Corpus: sample:3000@arrayexpress-ena, a seeded draw from the 20,693 studies BioStudies records an ENA link for. Surveyed: 2026-09-15; the nine studies whose SDRF is split across two files re-asked on 2026-10-27, once the route read both.

Resolved	2,324 (77.5%)
Empty	674
Failed	2
Unique file URIs	197,194

Median 16 files per resolved study; mean 85, which is the distribution's own warning that a handful of studies carry thousands.

Split by provenance, the same way every other ArrayExpress route splits:

Prefix	Studies	Names fastq URIs
E-MTAB	1,398	1,385	99%
E-GEOD	1,550	887	57%
E-ERAD	48	48	100%

Unlike the BioSample column, which is absent from every E-GEOD SDRF, fastq URIs survive the import into ArrayExpress a little over half the time. That is the difference between a column GEO has no equivalent of and one it does.

Both failures are a 404 on the SDRF itself, and neither is transient. They are E-MTAB-14460 and E-MTAB-15118, the two of the five studies that register no files at all that this draw happened to include. There were 11 until 2026-10-27. The other nine are E-GEOD imports (E-GEOD-15593 and eight more) whose SDRF is split across two files, .hyb.sdrf.txt and .seq.sdrf.txt, with none by the usual name. The route now reads the files the IDF names: eight of the nine name fastq URIs and resolve, and E-GEOD-30583 names its runs but no fastq. The same two fail on the SDRF's BioSample route; both are read from data/surveys/sample_3000@arrayexpress-ena/.