Single-Cell Biology · arcinstitute
Stack-CellxGene45M
A curated collection of 45 million single-cell transcriptomic profiles for model alignment and cross-study analysis.
What Stack-CellxGene45M is
Stack-CellxGene45M is a curated collection of approximately 45 million single-cell RNA sequencing profiles drawn from the CellxGene corpus, developed by the Arc Institute. It provides standardised cellular data designed for applications across biology, genomics, and medicine.
The dataset was compiled to align the Stack model following its pretraining on full human scBaseCount. It provides structured transcriptomic profiles intended for in-context learning and cross-study perturbation analysis.
How it was made
The dataset was curated from the broader CellxGene corpus by applying specific dataset-level filtering thresholds. To be included, an individual dataset was required to have at least 50,000 cells and span at least 5 distinct donors.
Cell identities were standardised by heuristically identifying author-annotated, coarse-grained cell type labels across datasets and storing them in the metadata field adata.obs["author_cell_type"].
Key facts
- Total size
- ~45 million cells
- Source corpus
- CellxGene
- Cell threshold per dataset
- ≥ 50,000 cells
- Donor threshold per dataset
- ≥ 5 donors
- Cell type metadata field
- adata.obs["author_cell_type"]
What you can use it for
- Stack Model Alignment
- Fine-tuning and aligning the Stack foundation model after its initial pretraining on full human scBaseCount.
- In-Context Learning
- Training and evaluating single-cell machine learning models capable of in-context learning across diverse cell states.
- Cross-Study Perturbation Analysis
- Analysing and comparing cellular perturbation responses across multi-donor, large-scale single-cell transcriptomic studies.
Explore the data
No queryable preview
This dataset is not available in a queryable form — usually an unusual file format or a repository too large to convert. The original source may still be readable.Strengths and limits
Good at
- Comprises roughly 45 million standardised single-cell transcriptomic profiles.
- Enforces high data density and diversity thresholds of at least 50,000 cells and 5 donors per constituent study.
- Provides harmonised coarse-grained cell type metadata in a consistent AnnData observation field.
Watch out for
- Excludes smaller experimental datasets containing fewer than 50,000 cells or fewer than 5 donors.
- Cell type annotations rely on heuristic identification and transfer of author-provided coarse-grained labels rather than exhaustive manual re-annotation.
Questions
Can I look inside Stack-CellxGene45M without downloading it?
Not here: it is not available in a queryable form. Its description and statistics are still shown.
What licence is Stack-CellxGene45M under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
What is the primary purpose of Stack-CellxGene45M?
It was curated to align the Stack model after pretraining on full human scBaseCount, and to support in-context learning and cross-study perturbation analysis.
What filtering criteria were applied to the CellxGene corpus?
Studies were included only if they contained at least 50,000 cells and data from at least 5 donors.
How are cell types recorded in the dataset?
Coarse-grained cell type labels provided by the original authors were heuristically detected and assigned to adata.obs["author_cell_type"].
Written from Stack-CellxGene45M's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.