README.md

metadata

license: mit

Data Card: Sepsis vs. SIRS Point-of-Care Biomarker Whole-Blood Microarray Dataset (GSE236713)

Summary

Expression + sample metadata + feature metadata for GSE236713, a multi-center UK study identifying transcriptional mRNA biomarkers to discriminate sepsis from SIRS in adult ICU patients, profiled on the Agilent SurePrint G3 Human GE v2 8x60K Microarray (GPL17077). Blood was sampled at up to four timepoints (Day 1, Day 2, Day 5, and ICU discharge) for patient groups, and once for healthy controls.

Source accession

Accession	N (samples)	N (patients)	Platform	Retrieval source
GSE236713	447	194	Agilent SurePrint G3 Human GE v2 8x60K, probe-name version (GPL17077)	NCBI GEO

Cohort breakdown: 30 healthy controls, 93 SIRS (Out-of-Hospital Cardiac Arrest, OOHCA), 124 abdominal sepsis, 200 pulmonary sepsis.

Files

• sample_metadata.parquet — one row per sample: sample_id, patient_id, timepoint (Day: 1/Day: 2/Day: 5/Day: Discharge), outcome (Died/Survived), disease (Control/SIRS/Sepsis), disease_type (Healthy/OOHCA/Abdominal/Pulmonary), sex
• feature_metadata.parquet — one row per probe (feature_id): Agilent's own platform annotation table, retrieved directly from GEO's GPL17077 record (no dedicated Bioconductor annotation package exists for this array design — confirmed via multiple Bioconductor support threads spanning several years, none reporting a resolution)
• expression.parquet — single file, long format: feature_id, sample_id, value

Sample metadata field notes

• Not every patient has all four timepoints. Of 194 patients: 55 have 1 sample, 48 have 2, 68 have 3, 23 have 4. Timepoints were missed due to patient death, ICU discharge, or other events, per the accession's design description. Healthy controls were sampled once and recorded as Day: 1 only.
• outcome is Died (91 samples) or Survived (356 samples).
• disease/disease_type together describe both severity tier and clinical subtype — disease_type is not a further breakdown of disease independent of it; each disease value maps to exactly one disease_type (Control→Healthy, SIRS→OOHCA, Sepsis→{Abdominal, Pulmonary}).

Expression value processing

Per the accession, expression values are normalized signal intensity. Independently confirmed as already log-scale: the standard log-detection heuristic (99th percentile and range of values) returned FALSE, indicating no further transformation was needed or applied. Values in expression.parquet are exactly as deposited.

Provenance / reproducibility

See pull_gse236713.R