Datasets:

cmatkhan
/

gains

Combined, harmonized expression + sample metadata + feature metadata for four ArrayExpress accessions comprising the Genomic Advances in Sepsis (GAinS) microarray cohorts used to derive and validate the Sepsis Response Signature (SRS1/SRS2).

All four accessions were profiled on Illumina HumanHT-12 v4, array design A-MEXP-2210. The sample metadata from each accession has been extracted and harmonized. This may be reconstructed using the script pull_gains.R, also provided here.

Source accessions

Accession	Nickname	Disease	N (SDRF rows)	Author
E-MTAB-4421	Derivation	CAP only	270	Davenport (Oxford)
E-MTAB-4451	Validation	CAP only	114	Davenport (Oxford)
E-MTAB-5273	Discovery	CAP + FP + non-septic controls	237	Burnham (Oxford)
E-MTAB-5274	Validation	CAP + FP	108	Burnham (Oxford)

Files

sample_metadata.parquet — one row per sample: sample_id, accession, age, sex, survived_28d, srs_group, disease_state
feature_metadata.parquet — one row per probe (IlluminaID), union across all four accessions: annotation from illuminaHumanv4.db (SYMBOL, ENTREZID, ENSEMBL, GENENAME, UNIPROT) plus reannotation QC fields from the package's internal ExtraInfo table (ProbeQuality, CodingZone, GenomicLocation, SecondMatches, OtherGenomicMatches, RepeatMask, OverlappingSNP, ProbeSequence)
expression/ — partitioned parquet dataset (by accession), long format: IlluminaID, sample_id, accession, value (VSN-normalized log2 expression, as deposited by submitters)

Notes

In the accession pages the IDF file sometimes contains notes on sample QC failures. For instance, E-MTAB-4421 says there are 270 samples but only 265 are provided. In the IDF, it notes that 5 samples failed QC. See the IDF from the accession for more info if curious.
disease_state is derived from the sample_id prefix, not an explicit SDRF field in most accessions: CAP* → "CAP", FP* → "FP", CON* → "normal". Cross-checked against each accession's IDF Experiment Description text where available.
Serial/repeated sampling. E-MTAB-5273 and E-MTAB-5274 seem to have some repeated measures on a subset of patients. These subjects have sample_id with suffixes, eg CAP0003.B.1, CAP0003.B.3, CAP0003.B.5. This is not parsed into a separate day/timepoint column currently.

Provenance / reproducibility

Data retrieved with ArrayExpress::getAE() with probe annotation from Bioconductor's illuminaHumanv4.db. See pull_gains.R.