Ghana sepsis teaching data
These two files support part 4, Basic figures with ggplot2. They come from the source data that the authors published with:
Chenoweth JG, Colantuoni C, Striegel DA, et al. Gene expression signatures in blood from a West African sepsis cohort define host response phenotypes. Nature Communications 15, 4606 (2024). https://doi.org/10.1038/s41467-024-48821-0
Source data: Zenodo record 10916993, source_data_1.xlsx and source_data_7.xlsx (https://doi.org/10.5281/zenodo.10916993), licensed CC BY 4.0. Credit the paper and the Zenodo record when you reuse these files.
The raw RNA sequencing data and patient-level clinical records are in dbGaP (phs003608) under controlled access. These files contain only the de-identified values the authors shared to draw their figures.
ghana_sepsis_samples.csv
One row per blood sample (566 rows). Built from the demographics, pca-ghana, and patterns-ghana sheets of source_data_1.xlsx, joined on the sample ID.
| Column | Meaning |
|---|---|
sample_id |
De-identified sample ID from the source file |
patient_id |
Person ID: the part of sample_id before the underscore |
outcome |
Healthy (healthy donor), Survived, or Died (28-day outcome) |
visit |
H for healthy donors; otherwise 0h, 6h, 24h, 48h, 72h after enrolment, then 28d, 6m, 12m |
age, sex |
As recorded in the source file |
PC1, PC2 |
Principal component scores from the authors’ PCA of gene expression |
P1 to P30 |
CoGAPS pattern values (0 to 1) for the 30 patterns in the paper |
Changes from the source:
- Seven samples from three patients with unknown 28-day outcome were removed, as in the paper’s outcome comparisons. That leaves 42 healthy donors and 117 patients (63 survived, 54 died by day 28).
- Source visit codes
00h,06h,06mwere shortened to0h,6h,6m, and healthy donors’ visit was recoded from00htoH. Pattern1, …,Pattern30were renamedP1, …,P30. PC scores were rounded to 3 decimal places and pattern values to 4.- Only PC1 and PC2 are kept. The file has no percentage of variance explained, so the lesson’s axis labels don’t show one.
Healthy donors gave one sample. Patients gave between one and eight samples (median five). Patients who died have no samples after 72 hours, and only some survivors returned at 28 days (41), 6 months (25), and 12 months (14).
ghana_sepsis_heatmaps.csv
The values behind Figure 2a–f, in long format: one row per heatmap cell (1,530 rows). Built from the pattern-z, pattern-auroc, pattern-cibersort, pattern-hema, pattern-chemvitals, and pattern-proteins sheets of source_data_7.xlsx.
| Column | Meaning |
|---|---|
panel |
CoGAPS z-score, AUROC, CIBERSORT, Hematology, Chemistry and vitals, or Proteins |
pattern |
Pattern number, 1 to 30 |
group |
For the z-score panel: Healthy, Survived, or Died. Otherwise the panel name |
column |
Heatmap column label (visit, cell type, lab test, vital sign, or protein) |
value |
Cell value, rounded to 3 decimal places |
What the panels contain, as described in the paper:
- CoGAPS z-score: the mean pattern value in each group, standardized across each row. Recomputing this from
ghana_sepsis_samples.csv(group means, thenscale()within each pattern) reproduces the shared values within rounding precision. - AUROC: how well each pattern separated 28-day survivors from non-survivors at each of the first five visits.
- CIBERSORT, Hematology, Chemistry and vitals, Proteins: Spearman correlations between pattern values and estimated cell proportions, blood counts, chemistries and vital signs, or plasma proteins.
Changes from the source: pattern labels were standardized to numbers, visit codes shortened as above, and some column labels spelled out (for example HR to Heart rate, CD56 Bright NK to NK cells).
The paper orders rows by hierarchical clustering. Clustering the z-scores with hclust() (complete linkage) gives the same six groups of patterns, but the top-to-bottom order can differ. The lesson sets the published order directly so the heatmap matches Figure 2.