scGPT

scGPT is a generative foundation model for single-cell multi-omics, pretrained on a corpus of over 33 million human cells, developed by the Wang lab (University Health Network and University of Toronto). It is peer-reviewed in Nature Methods (Cui, Wang, Maan et al., "scGPT: toward building a foundation model for single-cell multi-omics using generative AI", Nature Methods, 2024, doi:10.1038/s41592-024-02201-0; the repository's own citation still points to the earlier 2023 bioRxiv preprint, which this description notes rather than overriding). scGPT ships a model zoo of pretrained checkpoints (whole-human, brain, blood, heart, lung, kidney, pan-cancer) and is used for cell-type annotation, perturbation prediction, and multi-omic integration, standard single-cell genomics research workflows also listed on the Chan Zuckerberg Initiative's Virtual Cells Platform. It is a general single-cell representation model rather than a variant-effect predictor, similar in kind to this registry's existing Geneformer entry.

Licence note: MIT, confirmed directly against the repository's own raw LICENSE file (not GitHub's derived badge alone).

Registry note on model_card_uri: this points to the original authors' model zoo in github.com/bowang-lab/scGPT, which describes each pretrained checkpoint and links its download. It is not a GA4GH Genomic AI Model Card v0.6.0-formatted document. A Hugging Face copy (tdc/scGPT) is a third-party conversion by Therapeutics Data Commons, not a release by the original authors. No model has yet published a card in the GA4GH v0.6.0 format, since it is a newly drafted specification. The primary source repository remains github.com/bowang-lab/scGPT.