Genomics & Single-Cell · University of Toronto
scGPT
A foundation model trained on 33 million human cells for single-cell annotation and perturbation prediction.
What scGPT is
scGPT is a single-cell biology foundation model developed by the University of Toronto. It addresses analytical challenges in single-cell genomics by representing an individual cell's gene expression profile conceptually as a sentence composed of genes.
Pre-trained on 33 million human cells, the model provides transferable representations for downstream single-cell analysis tasks. It serves as a general backbone for workflows involving cell classification, experimental response forecasting, and data harmonisation.
How it works
scGPT processes single-cell transcriptomic profiles by treating a cell's expression measurements as an ordered sentence of genes. The model is pre-trained across 33 million human cells to learn gene-gene relationships and cellular states before being transferred to task-specific downstream objectives.
Key facts
- Version
- 1.0
- Training data
- 33 million human cells
- Domain
- Genomics
- Modalities
- Single-cell
- Memory requirement
- 40 GB GPU memory
What you can use it for
- Cell Type Annotation
- Assign labels to individual cells from an expression matrix, returning per-cell predictions accompanied by confidence scores.
- Perturbation Response Prediction
- Model and forecast how single cells respond transcriptionally to experimental perturbations.
- Batch Integration
- Harmonise expression matrices across different experimental batches using transferred cellular representations.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Not runnable yet
This model isn't available to run right now.Run this model
scGPT
Strengths and limits
Good at
- Pre-trained on a broad corpus of 33 million human single cells.
- Transfers effectively across multiple core tasks including annotation, perturbation prediction, and batch integration.
- Supplies confidence estimates alongside per-cell annotation predictions.
Watch out for
- Requires high-end computational resources, needing 40 GB of GPU memory to run.
- Restricted to expression data supplied in h5ad format via URL.
Inputs and outputs
You provide
Expression matrix (h5ad URL)
Taskoptional
You get back
Per-cell predictions with confidence.
Licence
Commercial use permitted · MIT
The MIT licence allows use in commercial research and products.
Questions
Can I use scGPT commercially?
Commercial use permitted. The MIT licence allows use in commercial research and products.
How do I run scGPT without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on 2× A100, typically in 5.0min, and a run costs about 2,100 credits. Visitors without an account get 1,000 credits a month.
What inputs does scGPT take?
The model takes an expression matrix provided via an h5ad URL and a selected target task.
What outputs does the model generate?
It outputs tabular per-cell prediction results containing annotations and associated confidence scores.
Which single-cell tasks are supported?
The model supports cell type annotation, perturbation response prediction, and batch integration.
Written from scGPT's own documentation. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.