---
license: mit
---
# Data Card: Hyperinflammatory/Hypoinflammatory Sepsis Phenotype Whole-Blood RNA-seq Dataset (GSE236892)

## Summary

Expression + sample metadata + feature metadata for GSE236892, a
whole-blood RNA-seq study comparing the hyperinflammatory and
hypoinflammatory molecular phenotypes of sepsis — two phenotypes
previously identified by latent class analysis (LCA) across multiple
cohorts, with divergent clinical outcomes and treatment responses. The
study reports 5,755 differentially expressed genes (31% of genes tested)
between phenotypes: hyperinflammatory patients showed elevated innate
immune response gene expression, hypoinflammatory patients showed elevated
adaptive/T-cell response gene expression. The study also reports
concordance with other previously described sepsis/ARDS molecular
subtypes (SRS1-2, MARS1-4, reactive/uninflamed) and a plasma metagenomic
analysis (not part of this GEO deposit).

## Source accession

| Accession | N (patients) | Retrieval source |
|---|---|---|
| [GSE236892](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE236892) | 113 hypoinflammatory + 76 hyperinflammatory = 189 | NCBI GEO |

## Files

- `sample_metadata.parquet` — one row per sample: `sample_id`, `age` (integer; see note on `imputed_age`), `sex`, `lca_label` (factor, `Hypo`/`Hyper` — the LCA-assigned phenotype), `imputed_age` (logical), `age_scaled` (numeric, standardized)
- `feature_metadata.parquet` — one row per gene (`feature_id`, Ensembl gene ID, version suffix stripped): gene coordinates from a GENCODE v49 GTF-derived `TxDb`, gene symbol/name/type from `org.Hs.eg.db`, plus two derived flags: `autosomal_protein_coding` and `hemoglobin_related` (see notes)
- `expression.parquet` — single file, long format: `feature_id`, `sample_id`, `value`

## Sample metadata field notes

- **`lca_label`** is the phenotype variable used in this dataset —
  patients were previously assigned Hyper/Hypo status via latent class
  analysis in the source study, not derived here. A separate
  `sepsis_group` value was parsed from the sample `title` field during
  retrieval but was **not** carried into the final stored metadata;
  `lca_label` is the field to use.
- **`age`**: some source values were recorded as `"90+"` rather than a
  specific integer. These were coerced to `90` and flagged via
  `imputed_age = TRUE` — treat `age` as a floor, not an exact value, for
  any sample where `imputed_age` is `TRUE`.
- **`age_scaled`** is `age` standardized (mean-centered, unit variance) —
  provided for direct use as a model covariate; not a raw clinical value.

## Feature metadata notes

- Gene coordinates/annotation come from GENCODE v49.
  **Genes present in the count matrix but absent from the GENCODE v49
  `TxDb`**: There are 378 that did not map btwn the cnt matrix and the
  v49 annotations. those are assigned to chrUn in the feature_meta.
- **`autosomal_protein_coding`** flags genes that are protein-coding
  (per `GENETYPE`) and on an autosome (chr1–22) — useful as a standard
  filtering criterion for downstream analyses, computed here rather than
  left for every user to redefine themselves.
- **`hemoglobin_related`** flags the 12 major hemoglobin genes/pseudogenes
  (`HBA1`, `HBA2`, `HBB`, `HBBP1`, `HBD`, `HBE1`, `HBG1`, `HBG2`, `HBM`,
  `HBQ1`, `HBZ`, `HBZP1`) — relevant for whole-blood RNA-seq, where
  globin transcript abundance is a common QC/filtering consideration.

## Expression value processing

**Raw, integer gene-level counts**, as provided in the submitter's
supplementary count matrix. No normalization or filtering applied here.

## Provenance / reproducibility

See [pull_gse236892.R](pull_gse236892.R)