Update README.md

README.md

@@ -1,3 +1,53 @@
 ---
 license: mit
 ---
+# Data Card: Septic Shock Early Therapy Response Whole-Blood RNA-seq Dataset (GSE110487)
+
+## Summary
+
+Expression + sample metadata + feature metadata for GSE110487, a
+whole-blood RNA-seq study of septic shock patients profiled for
+transcriptomic differences between early responders and non-responders to
+supportive hemodynamic therapy, assessed by SOFA score improvement within
+the first 48 hours of ICU admission.
+
+## Source accession
+
+| Accession | N (samples) | N (patients) | Platform | Retrieval source |
+|---|---|---|---|---|
+| GSE110487 | 62 | 31 (17 responders, 14 non-responders) | Illumina Genome Analyzer IIx (GPL10999, 6 samples) + Illumina HiSeq 2500 (GPL16791, 56 samples) | NCBI GEO |
+
+Each patient contributed 2 samples: T1 (within 16 hours of ICU admission)
+and T2 (within 48 hours). This SuperSeries splits its 62 samples across
+two sequencing platforms — both were combined to build sample metadata,
+since NCBI's RNA-seq reprocessing counts file pools both platforms into
+one matrix.
+
+## Files
+
+- `sample_metadata.parquet` — one row per sample: `sample_id`, `patient_id`, `response_to_treatment` (logical; `TRUE` = Responder), `timepoint` (`T1`/`T2`)
+- `feature_metadata.parquet` — one row per gene (`GeneID`, NCBI Entrez Gene ID): NCBI's GRCh38.p13 gene annotation file (`Human.GRCh38.p13.annot.tsv.gz`), matched to the same reprocessing build as the expression data
+- `expression.parquet` — single file, long format: `GeneID`, `sample_id`, `value`
+
+## Sample metadata field notes
+
+- **`response_to_treatment`** is a per-patient classification (Responder
+  vs. Not Responder, defined by SOFA score improvement within 48 hours),
+  constant across a patient's T1/T2 samples — unlike some other datasets
+  in this collection, this label does not vary by timepoint.
+- **`timepoint`** is T1 (within 16h of ICU admission) or T2 (within 48h).
+  28 T1 samples, 34 T2 samples — not perfectly balanced, worth noting if
+  used in a paired analysis.
+- All 31 patients have exactly 2 samples (one T1, one T2); no missing
+  timepoints.
+- No age or sex data available in the accession's sample-level metadata.
+
+## Expression value processing
+
+**Raw, un-normalized gene-level counts**, from NCBI's own RNA-seq
+reprocessing pipeline (GRCh38.p13 realignment).
+No filtering or normalization has been applied.
+
+## Provenance / reproducibility
+
+See [pull_gse110487.R](pull_gse110487.R).