Shorten dataset card to sources, preprocessing and use

README.md

@@ -20,25 +20,47 @@ configs:
 
 # CausalBench: K562 and RPE1
 
-Both filtered datasets in one place, prepared from Replogle et al.’s Perturb-seq
-count matrices using CausalBench’s preprocessing. These files are unchanged
-from the earlier [K562](https://huggingface.co/datasets/huggingbahl21/causalbench-k562)
-and [RPE1](https://huggingface.co/datasets/huggingbahl21/causalbench-rpe1) uploads.
+Filtered single-cell CRISPRi perturbation data from
+[Replogle et al. (2022)](https://doi.org/10.25452/figshare.plus.20029387.v1),
+prepared following [CausalBench](https://www.nature.com/articles/s42003-025-07764-y).
 
-| Dataset | Cells | Gene columns |
-| --- | ---: | ---: |
-| K562 | 162,751 | 622 |
-| RPE1 | 162,733 | 383 |
+| Dataset | Cells | Genes | File |
+| --- | ---: | ---: | --- |
+| K562 | 162,751 | 622 | `data/prepared/causalbench_k562.tsv.gz` |
+| RPE1 | 162,733 | 383 | `data/prepared/causalbench_rpe1.tsv.gz` |
 
 Rows are cells. Gene columns use Ensembl IDs and contain normalized,
-log-transformed expression. The extra `intervention` column names the target:
-`non-targeting` means control; `excluded` means an unsupported target whose row
-was kept. **Excluded cells are not controls.** No ground-truth causal graph or
-training/test split is included; each subset contains the full filtered data.
-
-## Use the data
-
-Install `pandas` and `huggingface_hub`, then:
+log-transformed expression. Each table also has an `intervention` column:
+
+- A gene ID identifies the targeted gene.
+- `non-targeting` identifies control cells.
+- `excluded` identifies unsupported targets. These rows remain in the data
+  and **must not be treated as controls**.
+
+## Sources and preprocessing
+
+Inputs: [K562 counts](https://ndownloader.figshare.com/files/35773219),
+[RPE1 counts](https://ndownloader.figshare.com/files/35775606), and the
+[perturbation summary workbook](https://ars.els-cdn.com/content/image/1-s2.0-S0092867422005979-mmc2.xlsx)
+(sheets `TabB_K562_day6_summary_stat` and `TabC_RPE1_summary_statistic`).
+
+1. Keep controls and strong perturbations: >50 differentially expressed genes,
+   knockdown ≤−0.3, and >25 filtered cells in the summary.
+2. For measured targets, retain perturbed cells with target expression at or
+   below the control 10th percentile.
+3. Remove zero-total cells. Normalize each cell’s total to the retained-cell
+   median using **all original genes**, then apply natural `log1p`.
+4. Keep measured genes targeted in >100 retained cells. Preserve gene order
+   and relabel unsupported targets `excluded` without removing their rows.
+
+[`upstream/preprocessing.py`](upstream/preprocessing.py) is copied byte for byte
+from [CausalBench commit 1a2143c](https://github.com/causalbench/causalbench/blob/1a2143cffdc85f835b41ce8d52034be1bf903e71/causalscbench/data_access/utils/preprocessing.py).
+[`prepare.py`](prepare.py) is our download/export wrapper. K562 was prepared
+with the original function; RPE1 used a memory-saving implementation of the
+same rules. Reconstructing both datasets from the raw inputs reproduced all
+saved expression values and ordered labels exactly.
+
+## Load a table
 
 ```python
 import pandas as pd
@@ -53,56 +75,18 @@ expression = pd.read_csv(path, sep="\t")
 interventions = expression.pop("intervention")
 ```
 
-## Preparation
-
-```text
-data/raw/ → prepare.py → data/prepared/
-```
-
-The inputs are the published [K562](https://ndownloader.figshare.com/files/35773219)
-and [RPE1](https://ndownloader.figshare.com/files/35775606) count matrices and the
-[perturbation summary workbook](https://ars.els-cdn.com/content/image/1-s2.0-S0092867422005979-mmc2.xlsx).
-The download links and sheet names are in [`prepare.py`](prepare.py).
-
-CausalBench keeps controls and strong perturbations (>50 DEGs, knockdown ≤−0.3,
->25 cells), filters measured targets at the control 10th percentile, removes
-zero-total cells, normalizes totals across all original genes to the retained-cell
-median, and applies `log1p`. It then keeps genes targeted in >100 retained cells
-and labels unsupported targets `excluded`.
-
-[`upstream/preprocessing.py`](upstream/preprocessing.py) is copied **unchanged,
-byte for byte**, from [CausalBench commit 1a2143c](https://github.com/causalbench/causalbench/blob/1a2143cffdc85f835b41ce8d52034be1bf903e71/causalscbench/data_access/utils/preprocessing.py).
-`prepare.py` is our small download/export wrapper, not copied upstream code.
-The existing K562 file used the original function; RPE1 used an equivalent
-memory-saving implementation. New rebuilds call the original function directly.
-
-To rebuild locally:
-
-```bash
-uv sync --locked
-uv run prepare.py --output-dir data/rebuilt
-```
-
-The Hub stores prepared TSVs and the code. Raw inputs stay local. Rebuilding
-needs substantial RAM; the original function loads the full matrix. The summary
-workbook’s URL returned HTTP 403 when checked, so you may need to place an
-existing copy at `data/raw/summary_stats.xlsx`.
-
-## Saved checks
-
-[`verification.json`](verification.json), previously called `results.json`,
-records file fingerprints, dimensions, and data comparisons—not model scores
-or CausalBench benchmark results. On 1 November 2026, both raw files matched
-publisher checksums, reconstructed NPZ values and ordered labels matched exactly,
-and both TSVs matched the earlier Hub uploads byte for byte.
+## Notes for downstream use
 
-The full inputs were reconstructed in chunks; the original function was checked
-on a small fixture rather than rerun on the full datasets. TSV decimal conversion
-introduced differences of at most about 0.00000024. The summary workbook was
-checked locally; a fresh remote copy could not be verified.
+These are filtered expression values, not raw counts. The target-expression
+filter selects cells with lower target expression, so knockdown strength in
+these tables is partly determined by that selection. The tables mix control
+and perturbed cells; separating the label column does not make them observational.
+No complete ground-truth causal graph or train/test split is provided.
+Perturbation effects can include indirect effects and do not establish direct edges.
 
-## Credit
+[`verification.json`](verification.json) records file checksums and data checks,
+not model performance or benchmark results. The local summary workbook was
+recorded, but its publisher URL returned HTTP 403 during verification.
 
-Data: [Replogle et al. (2022)](https://doi.org/10.25452/figshare.plus.20029387.v1),
-CC BY 4.0. Preprocessing: [CausalBench / Chevalley et al. (2025)](https://www.nature.com/articles/s42003-025-07764-y).
-The copied code retains its original attribution and [Apache 2.0 license](upstream/LICENSE.txt).
+Data: **CC BY 4.0**, credited to Replogle et al. Copied CausalBench code retains
+its original attribution and [Apache 2.0 license](upstream/LICENSE.txt).