Cytokine Differential Expression · tahoebio
parse-de-rhaister
Differential-expression statistics for 12 donors, 18 immune cell types, and 90 cytokines from a Parse PBMC screen.
What parse-de-rhaister is
parse-de-rhaister is a dataset of differential-expression summary statistics published by tahoebio. Derived from the Arc Institute ST-HVG-Parse single-cell dataset, it documents peripheral blood mononuclear cell (PBMC) responses to cytokine perturbations across 12 human donors, 18 immune cell types, and 90 cytokines.
The dataset compiles pre-computed differential-expression metrics—such as per-gene fold changes, non-parametric test statistics, and linear pseudobulk deltas—that were used to train the Rhaister perturbation model. By converting raw single-cell counts into standardised long- and wide-format tables, it allows researchers to study cell-state responses directly without processing raw sequencing reads.
How it was made
The dataset calculates statistics by evaluating cellular contexts, defined by donor and cell type pairs, against individual cytokine treatments and untreated controls from the Arc Institute ST-HVG-Parse screen. The primary table, pdex, stores long-format per-gene differential expression across roughly 774 million rows, including log2 fold changes, percentage changes, Mann–Whitney U test statistics, p-values, Benjamini–Hochberg false discovery rates (FDR), and reference and target cell counts.
A secondary wide-format table, cell_eval, provides pseudobulk expression deltas across approximately 40,000 gene columns, calculated as mean treated expression minus mean control expression on a linear normalised scale for each cell type, donor, and cytokine combination. The repository also provides dataset definition files with compositional split criteria, which define composite contexts as donor-cell type pairs and hold out specific cytokine-context combinations while retaining individual contexts and cytokines in the training set.
Key facts
- Publisher
- tahoebio
- Source dataset
- Arc Institute ST-HVG-Parse
- Experimental design
- 12 donors × 18 immune cell types × 90 cytokines
- Table pdex size
- ~774 million rows (long format)
- Table cell_eval dimensions
- One row per context-cytokine pair across ~40,000 gene columns
- Statistical metrics
- Log2 fold change, Mann–Whitney U test, Benjamini–Hochberg FDR
- Split configurations
- Compositional splits (split_0 to split_4) and donor splits
What you can use it for
- Perturbation model training
- Supplying pseudobulk linear deltas and differential-expression statistics to train predictive models of immune response, such as Rhaister.
- Immune cell cytokine profiling
- Evaluating how specific immune subsets vary in gene expression changes, fold changes, and significance under 90 different cytokine conditions.
- Generalisation benchmarking
- Testing model predictions on unseen donor, cell type, and cytokine pairings using the pre-defined compositional holdout splits.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from tahoebio/parse-de-rhaister at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Pre-computes differential-expression values and significance tests across approximately 774 million gene-perturbation combinations.
- Provides wide-format pseudobulk deltas for roughly 40,000 genes alongside cell-level group statistics.
- Includes formal compositional train-test split definitions designed for evaluating perturbation models.
Watch out for
- Contains summary-level statistics and pseudobulk deltas rather than single-cell level expression matrices or raw sequencing reads.
- Log2 fold change values evaluate to negative infinity when target expression is zero.
- Data coverage is restricted to the 12 donors, 18 immune cell types, and 90 cytokines present in the source screen.
Questions
Can I look inside parse-de-rhaister without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is parse-de-rhaister under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
What is the difference between the pdex and cell_eval tables?
The pdex table is in long format with one row per donor, cell type, cytokine, and gene, detailing statistical tests and fold changes. The cell_eval table is in wide format, listing pseudobulk linear deltas across approximately 40,000 gene columns for each cell type, donor, and cytokine.
What statistical test is used to calculate significance?
Significance is calculated using a two-sided Mann–Whitney U test comparing treated cells against control cells, with Benjamini–Hochberg adjustment to compute false discovery rates (FDR).
How are the compositional splits constructed?
The compositional splits group data by composite context (donor combined with cell type) and hold out specific cytokines only within selected contexts, ensuring every test context and cytokine has also been observed separately during training.
Written from parse-de-rhaister's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.