mindflow-cn/scDrugPerturb-Bench

Predict the biology, not only the distribution.

Many drug-perturbation benchmarks ask whether a predicted treated population resembles the observed post-perturbation expression distribution. That endpoint is important, but distributional similarity alone does not establish that a model recovered the biological mechanism emphasized by the experiment.

scDrugPerturb-Bench adds a complementary, mechanism-aware evaluation layer. For every benchmark, we read the associated paper and encode its key gene-level response claims as structured tests: which genes should change, in which direction, under which drug, dose, time, and cellular context. These paper-grounded tests are stored in test_case.json beside the control and treated H5AD matrices.

At a glance

Papers	Benchmarks	Perturbations	Cells	Paper-grounded targets	Core data
20	20	17	527,774	212	30.4 GiB

What makes it different?

Evaluation view	Distribution-centered perturbation benchmark	scDrugPerturb-Bench
Primary question	Does the predicted treated population match the observed distribution?	Does the prediction recover the specific biological response reported in the paper?
Supervision	Control and treated expression matrices	Matrices plus paper-grounded mechanistic test cases
Evaluation unit	Global or cell-population distribution	Named target genes and expected UP / DOWN / NS relations
Experimental context	Often implicit in metadata	Drug, dose, time, cell type, control, and comparison encoded explicitly

The two views are complementary. A useful perturbation model should reproduce the treated distribution and recover the response programs that make the perturbation biologically meaningful.

The mechanism contract: test_case.json

Each benchmark turns statements from the source paper into machine-readable biological assertions. For example, the Brefeldin A benchmark encodes a coordinated UPR/ER-stress response in HepaRG cells. The example below preserves the complete structure of the published test_case.json:

{
  "benchmark_id": "40766395_01",
  "sample_system": "",
  "tissue": "Liver",
  "source_type": "cell_line",
  "perturbation_type": "drug",
  "perturbation_name": "Brefeldin A",
  "default_cell_subset": "",
  "description": "HepaRG cells treated with Brefeldin A.",
  "smiles": "CC1CCCC=CC2CC(CC2C(C=CC(=O)O1)O)O",
  "test_cases": [
    {
      "test_id": "40766395_01_01",
      "target_genes": [
        "HSPA5",
        "GTF2F1",
        "GOLGA3",
        "SYVN1",
        "LMF2",
        "HYOU1",
        "SLC38A10",
        "ZNF598",
        "PDIA4",
        "SQSTM1",
        "DDIT3",
        "ATF4",
        "INHBE",
        "ARF4"
      ],
      "perturb_var": "dose",
      "control": "C",
      "dose_groups": ["0.01 uM", "0.1 uM"],
      "time_groups": ["24 h"],
      "relation": "UP",
      "cell_type": "HepaRG liver toxicology cell line"
    }
  ]
}

This makes evaluation explicit and auditable: a model can match a broad distribution while still missing a lineage switch, stress program, resistance mechanism, or cell-cycle response that the paper identifies as central.

Repository structure

README.md
benchmarks.csv
data/
  <benchmark_id>/
    control.h5ad
    ground_truth.h5ad
    test_case.json

File	Role
benchmarks.csv	One row per benchmark with publication and dataset provenance, species and cellular context, tissue and disease, sequencing platform, perturbation identity and SMILES, matched control, and dose/time design. Use it to discover relevant experiments and trace every benchmark back to its source study.
control.h5ad	Processed single-cell expression before perturbation or under the matched control condition.
ground_truth.h5ad	Observed single-cell expression after perturbation.
test_case.json	Paper-grounded target genes, expected relation, dose/time groups, cell context, and test identifiers.

X stores the processed expression matrix. Observation and variable metadata are retained, together with a counts layer where available. Always align matrices by var_names, not by raw column position.

Quick start

import json
from pathlib import Path

import anndata as ad

benchmark_dir = Path("data/40766395_01")
control = ad.read_h5ad(benchmark_dir / "control.h5ad")
treated = ad.read_h5ad(benchmark_dir / "ground_truth.h5ad")
test_spec = json.loads((benchmark_dir / "test_case.json").read_text())

for case in test_spec["test_cases"]:
    print(case["test_id"], case["target_genes"], case["relation"])

Recommended evaluation has two parts:

1. Measure how well the predicted cells reproduce the observed treated distribution.
2. Test whether the predicted target-gene changes satisfy the paper-grounded relations in test_case.json.

Curated benchmarks

Rank	Benchmark	PMID	Perturbation	Tissue	Targets	Direction agreement
1	38937474_01	38937474	osimertinib	Lung	34	32/34 (94.1%)
2	38895265_01	38895265	Paclitaxel	Breast	18	18/18 (100.0%)
3	34591417_01	34591417	vemurafenib	Skin	14	14/14 (100.0%)
4	40766395_01	40766395	Brefeldin A	Liver	14	14/14 (100.0%)
5	37086265_01	37086265	etoposide	Lung	18	16/18 (88.9%)
6	33712615_01	33712615	erlotinib	Lung	19	15/19 (78.9%)
7	32846134_01	32846134	5-fluorouracil	Colon	9	9/9 (100.0%)
8	37732484_01	37732484	paclitaxel	Aorta	11	10/11 (90.9%)
9	36318267_01	36318267	estradiol	Breast	8	8/8 (100.0%)
10	36553506_01	36553506	panobinostat	Brain	12	10/12 (83.3%)
11	35410383_01	35410383	Fluorouracil	Breast	6	6/6 (100.0%)
12	41871169_01	41871169	panobinostat	B lymphoblast	6	6/6 (100.0%)
13	38652658_01	38652658	ispinesib	Brain	9	8/9 (88.9%)
14	36382181_01	36382181	enzalutamide	Prostate	9	7/9 (77.8%)
15	32094658_01	32094658	latrunculin A	Pancreas	4	4/4 (100.0%)
16	38272949_02	38272949	GW3965	Brain	4	4/4 (100.0%)
17	38589664_01	38589664	cisplatin	Stomach	3	3/3 (100.0%)
18	39803533_01	39803533	TCDD	Skin	3	3/3 (100.0%)
19	40166195_01	40166195	doxorubicin	Breast	3	3/3 (100.0%)
20	34857732_01	34857732	GSK126	Prostate	8	5/8 (62.5%)

The collection spans 17 drugs or small-molecule perturbations across cancer, differentiation, stress, senescence, neurobiology, vascular biology, and other cellular contexts. Each PMID appears only once, preventing multiple branches of the same paper from dominating the release.

Paper-to-data consistency

Across 212 within-benchmark unique target genes, 195 (92.0%) satisfy the encoded relation under a strict aggregate consistency check. The remaining 17 consist of 12 same-direction but sub-threshold effects, 3 paper-expected NS targets with observed changes, and 2 genuinely reversed aggregate directions.

For each target gene:

relative_effect = (treated_mean - control_mean) /
                  (abs(treated_mean) + abs(control_mean))

• relative_effect >= 0.1 is classified as UP.
• relative_effect <= -0.1 is classified as DOWN.
• Intermediate values are classified as NS.

This is a dataset-level consistency check, not a dose-stratified or time-stratified significance test. The expected relations remain paper-grounded assertions; the aggregate check documents how those assertions appear in the processed matrices.

License and responsible use

The repository uses license: other intentionally. It does not grant rights beyond those of the original studies. Users are responsible for checking source-dataset redistribution terms and for citing the corresponding papers listed in benchmarks.csv.

Citation

Please cite the relevant original paper for every benchmark used. PMID, DOI, paper title, journal, accession, experimental context, perturbation, dose design, and time design are provided in benchmarks.csv.