中文

Large perturbation datasets: what each record states about scale, system and provenance

By GlioState Editorial Team · Published · Last updated

Review pending Version 0.1.0 · status: draft · last expert review: not yet reviewed

Layer: General layer · Tissue scope: Cancer cell lines

Research question

  • Which large perturbation datasets and profiling platforms are in the library, and what do their records state about scale and cellular system?
  • Which statements are dataset-producer descriptions rather than independent evaluations?

Current understanding

Our reading of the library records: the large perturbation resources in the library are generated by a few producers and are described mainly by those producers. Genome-scale Perturb-seq in K562/RPE1 (peer reviewed), a drug-perturbation atlas of 100M cells (peer-reviewed Cell record plus a preprint record), genome-wide Perturb-seq from Xaira, a purpose-built Arc dataset for the 2026 Virtual Cell Challenge, and a probe-based profiling method are recorded. Records give scale and system but do not evaluate data quality independently. Certainty: high that records state this; not independently assessed.

Scope and how to read this page. Scope: library records only (title, venue, one-line summary and verification source); we have not read the full texts for this page, and every statement below is limited to what those records state. Preprints are marked and are not peer reviewed. This page makes no claim about glioma or other tumours. VC0040 VC0041 · preprint

Genome-scale Perturb-seq (peer reviewed). The Cell 2022 record is the genome-scale Perturb-seq study of Replogle et al., recorded as K562/RPE1 benchmark data. The data licence is CC BY 4.0, verified from the Figshare+ repository page (data licence, not the paper's licence). VC0040

Tahoe-100M. Two records exist: the bioRxiv preprint (100M cells, 1,100 drugs across 50 cancer cell lines; preprint) and the peer-reviewed Cell 2026 record. The data are CC0 1.0 on the Hugging Face repository page, verified; access was not gated. Certainty: high for scale as stated in the record; the two records describe the same resource and should not be counted as two datasets. VC0041 · preprint VC0042

X-Atlas/Orion (Xaira; preprint). Recorded as genome-wide Perturb-seq datasets generated with a scalable fix-cryopreserve platform. The Hugging Face repository lists CC BY-NC-SA 4.0 (free to access, but not an open licence in the sense of allowing commercial use), verified 2026-10-16. A related record describes scalable probe-based single-cell profiling for virtual-cell perturbation mapping (preprint). VC0045 · preprint VC0046 · preprint

A dataset built for a challenge. Arc describes how the 2026 Virtual Cell Challenge dataset was generated: CRISPRi Perturb-seq in six human cell lines with 10x Flex profiling, more than 10,000 perturbations and 100 non-targeting guides per line, curated to perturbations with at least 80% median on-target knockdown and subsampled to 400 cells per perturbation. This is an organisation statement, not an independent evaluation, and it reports no model results. The task is zero-shot: only unperturbed cells are given. The challenge record states three of the six lines are for validation and three held out. VC0143 VC0107

A preprint from a single-system study. A bioRxiv preprint (Norman/Dey labs) is recorded as CRISPRa Multiome Perturb-seq of 119 transcription factors and dual-guide Perturb-seq of 1,225 TF pairs in RPE-1, with a Transformer model (Mosaic) predicting TF-pair interaction residuals (r = 0.484) and nominating a triple-TF combination that the record calls validated. It is a perturbation resource in a non-cancer cell line, not a general benchmark, and is not peer reviewed. VC0139 · preprint

Evidence map

Primary studies & datasets

  • VC0040 Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq. — Replogle JM et al., Cell, 2022
  • VC0041 · preprint Tahoe-100M : A Giga-Scale Single-Cell Perturbation Atlas for Context-Dependent Gene Function and Cellular Modeling — Zhang J et al., bioRxiv, 2025
  • VC0042 Tahoe-100M: Mapping drug-induced molecular phenotypes at single-cell resolution. — Zhang J et al., Cell, 2026
  • VC0045 · preprint X-Atlas/Orion: Genome-wide Perturb-seq Datasets via a Scalable Fix-Cryopreserve Platform for Training Dose-Dependent Biological Foundation Models — Huang AC et al., bioRxiv, 2025
  • VC0143 Generating high-quality Perturb-seq data for the 2026 Virtual Cell Challenge — Arc Institute, Arc news, 2026
  • VC0139 · preprint Engineered regulatory states resolve context-dependent genetics in macular degeneration — Metzner E, Guruvayurappan K, Southard KM, Ardy RC, Dey KK, Norman TM., bioRxiv, 2026

Related models & tools

  • VC0046 · preprint Scalable probe-based single-cell transcriptional profiling for virtual cell perturbation mapping and synthetic biology phenotyping — Swinderman JT et al., bioRxiv, 2026
  • VC0107 Virtual Cell Challenge 2026 (zero-shot, 6 unseen cell lines) — Arc Institute, virtualcellchallenge.org, 2026

8 library items cited on this page; 4 are preprints (not peer reviewed).

Entities on this page

What remains uncertain

Records give scale and system as stated by dataset producers; the library does not record independent assessments of data quality, batch structure or reuse. VC0040 VC0045 · preprint VC0143

The cellular systems recorded are cell lines (K562, RPE1, 50 cancer cell lines, six human cell lines, RPE-1); the records do not establish how models trained or tested on these data behave in primary tissue or other contexts. VC0040 VC0041 · preprint VC0143 VC0139 · preprint

For research and education only. Not medical advice and not a recommendation of any diagnosis, treatment or product. Patients should consult their own physicians.