Perturb-Sapiens

# Perturb Sapiens: A Human Whole-Organism Atlas of Perturbed Cells ## Dataset Description **Perturb Sapiens** is an evolving database of AI-predicted single-cell perturbation responses, representing the **first human whole-organism atlas of perturbed cells**. Perturb Sapiens is generated using the post-trained **Stack model** ([Stack-Large-Aligned](https://huggingface.co/arcinstitute/Stack-Large-Aligned)), an in-context learning foundation model for single-cell biology. **Data Sources:** - **Prompt Data:** Parse/OpenProblems PBMC perturbation data - **Query Data:** [Tabula Sapiens](https://tabula-sapiens.sf.czbiohub.org/) ### Key Statistics (v1) | Metric | Value | |--------|-------| | Total Cells | 103+ million | | Human Tissues | 28 | | Cell Classes | 40 | | Drug Conditions | 111 (control: Dimethyl sulfoxide) | | Cytokine Conditions | 90 (control: PBS) | ## Dataset Structure ``` PerturbSapiensv1/ ├── drug/ │ ├── [drug_name].h5ad │ └── Dimethyl sulfoxide.h5ad # Control └── cytokine/ ├── [cytokine_name].h5ad └── PBS.h5ad # Control ``` ## Quick Start ```python import scanpy as sc # Load a perturbation condition (e.g., ADSF cytokine) adata = sc.read_h5ad('PerturbSapiensv1/cytokine/ADSF.h5ad') # Preprocessing (aligned with post-training settings) sc.experimental.pp.highly_variable_genes(adata, n_top_genes=4000) sc.pp.normalize_total(adata) sc.pp.log1p(adata) # Subset to high quality cells adata = adata[adata.obs['gen_logit']<2.5].copy() # Subset to highly variable genes adata = adata[:, adata.var['highly_variable']].copy() # Ready for downstream analysis! ``` ## Intended Uses - **Drug response prediction:** Explore how drugs affect gene expression across tissues - **Cytokine signaling studies:** Understand tissue-specific responses to immune signals - **Hypothesis generation:** Identify potential therapeutic targets ## Limitations - Predictions are generated by a foundation model and should be validated experimentally - Potentially less precision in predicting effects of weak perturbations ## Citation If you use Perturb Sapiens in your research, please cite: - Dong et al., 2026: Stack: In-context modeling of single-cell biology. Preprint. [Paper link](https://www.biorxiv.org/content/10.64898/2026.01.09.698608v1) ## Links - 🔗 **GitHub:** [ArcInstitute/stack](https://github.com/ArcInstitute/stack) - 🤗 **Model:** [Stack-Large-Aligned](https://huggingface.co/arcinstitute/Stack-Large-Aligned) ## License This dataset is released under the [CC BY 4.0 License](https://creativecommons.org/licenses/by/4.0/). ## Contact For questions or issues, please open an issue on the [GitHub repository](https://github.com/ArcInstitute/stack/issues).