I thought I had a good handle on the data model of GDC, but a recent query on the TARGET dataset revealed something I had not seen before.
TL;DR- what does it mean when multiple "cases" are associated with a file (NOT the opposite where we see multiple samples/aliquots coming from a single "case")? After downloading one of these files, it only contains a single column of data. I expected multiple columns corresponding to the multiple cases.
Specifically, I'm using the GDC API to download all count-based (HTSeq) RNA-seq files. I have a Python script, but when I encountered this issue I went to the GDC portal to confirm I was not crazy. There, I have the following filter parameters:
cases.project.program.name in ["TARGET"] and cases.project.project_id in ["TARGET-NBL"] and files.analysis.workflow_type in ["HTSeq - Counts"] and files.data_type in ["Gene Expression Quantification"]
This results in 161 files (as of this posting). In the table shown (just as in my script) I see that several of the files have 2 "cases". Here's a link to one such file:
That url seems to imply that this particular file would have data about TWO cases/biospecimens. Indeed, in my script, I see clinical data for two humans (e.g. they each have very distinct demographics). However, after downloading and uncompressing this file, I only see one column of data (the read counts for a single aliquot):
$ tar -zvtf gdc_download_20211109_200829.855799.tar.gz -rw-r--r-- 0/0 207 2021-11-09 15:08 MANIFEST.txt -rw-r--r-- 0/0 251351 2021-11-09 15:08 936e9a45-d879-4500-9571-7024e5dc053a/d5589fcc-c841-43d5-ae2a-ce5ee0914c7e.htseq_counts.txt.gz $ tar -xzf gdc_download_20211109_200829.855799.tar.gz $ zcat 936e9a45-d879-4500-9571-7024e5dc053a/d5589fcc-c841-43d5-ae2a-ce5ee0914c7e.htseq_counts.txt.gz | head ENSG00000000003.13 531 ENSG00000000005.5 12 ENSG00000000419.11 1857 ENSG00000000457.12 1243 ENSG00000000460.15 995 ENSG00000000938.11 83 ENSG00000000971.14 223 ENSG00000001036.12 908 ENSG00000001084.9 2536 ENSG00000001167.13 2885
Since this file allegedly contained two cases, yet only has a single column of data (with no header), I cannot properly associate these data with any clinical metadata, so it's not very useful. None of the UUIDs in the downloaded archive map to anything meaningful.
This same thing is found for other files associated with >1 biospecimens, so this is not an isolated issue.