tahoebio/replogle-nadig-de-rhaister
Replogle-Nadig CRISPR Screen Differential Expression Summary Statistics Differential-expression summary statistics for the Replogle-Nadig genome-scale CRISPR screen, which includes 4 cell lines and ~2,000 gene knockdowns. Used to train Rhaister, the data are per-gene fold changes, significance, and pseudobulk deltas summarized from raw single cells. The four cell lines are hepg2, jurkat, k562, and rpe1 and the perturbation is a CRISPR knockdown of a target gene. The repository… See the full description on the dataset page: https://huggingface.co/datasets/tahoebio/replogle-nadig-de-rhaister.
Replogle-Nadig CRISPR Screen Differential Expression Summary Statistics
Differential-expression summary statistics for the Replogle-Nadig genome-scale CRISPR screen, which includes 4 cell lines and ~2,000 gene knockdowns. Used to train Rhaister, the data are per-gene fold changes, significance, and pseudobulk deltas summarized from raw single cells. The four cell lines are hepg2, jurkat, k562, and rpe1 and the perturbation is a CRISPR knockdown of a target gene.
The repository contains two tables: pdex and cell_eval.
pdex per-gene differential expression (long format)
from datasets import load_dataset
pdex = load_dataset("tahoebio/replogle-nadig-de-rhaister", "pdex", streaming=True, split="train")One row per (cell line, gem_group, target KD, gene). ~1.5 B rows total.
cell_eval pseudobulk expression deltas (wide format)
cell_eval = load_dataset("tahoebio/replogle-nadig-de-rhaister", "cell_eval", streaming=True, split="train")One row per (cell line, gemgroup, target gene). Columns `cellline, gem_group, gene, then one column per measured gene (~6,546) holding the pseudobulk delta mean(perturbed) − mean(control)` on the linear (normalized) scale.
Definition & filtering criteria
definition/ holds the dataset definition (dataset.toml) and the holdout split TOMLs (split_0–split_3). It records the cell/treatment columns (cell_line, and gene as the treatment), the 2,000 highly-variable genes used for analysis (hvg_2k_genes.json), and the feature_names.json map from pdex's integer feature indices to gene symbols. The compositional split holds out a (cell line, gene) pair only when its cell line is in the holdout set and its target is in that cell line's held-out list, so each test cell line and target is still seen individually in training.
