X-Atlas/Pisces (25.6M Cell Multi-Context Perturb-seq)
Largest Perturb-seq dataset ever released: 25.6 million perturbed single-cell transcriptomes across 16 diverse biological contexts. Trained Xaira's X-Cell virtual cell model. Enables cross-context perturbation prediction and cell-type transfer learning.
Composite
94.4
Experimental validation
Wet-lab confirmed
Stages
Target ID
Modalities
genetic_perturbationsingle_cell
Task types
perturbation_predictioncontext_transferfoundation_model_training
Size
cells: 25,600,000
biological_contexts: 16
biological_contexts: 16
License
CC-BY-4.0
First release
2026-03
Last updated
2026-03
Official site
Leaderboard
→ leaderboard
Dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
X-Cell: A Virtual Cell Model Trained on X-Atlas/Pisces · · 2026 · paper · 10 citations
Flags
foundation_model_dataindustry_generatedlargest_in_class
Experts
—
Groups
—
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
5
coverage
5
maintenance
4
adoption
4
quality
5
accessibility
5
industry_relevance
5
Notes
Successor to X-Atlas/Orion. 16 diverse biological contexts enable robust cross-context generalization. Underlies X-Cell foundation model. Industry-scale data release.