Geneformer vs TranscriptFormer
Geneformer: Transformer pretrained on ~30M single-cell transcriptomes for network biology predictions. TranscriptFormer: CZI cross-species generative single-cell foundation model trained on 112M cells from 12 species.
Facts below come from each project's own website, repository and paper. We don't publish benchmark claims here; see the papers for evaluations. Shared tags: foundation-model.
| Field | Geneformer | TranscriptFormer |
|---|---|---|
| Pretraining corpus | ~30M (original) / ~104M transcriptomes | Up to 112M cells, 12 species |
| Architecture | Transformer | Generative transformer, 444M trainable params |
| Tasks | In silico perturbation, cell + gene classification (fine-tuning) | Cell-type classification, zero-shot disease state, cross-species annotation transfer |
| Species | Human | 12 species |
| License | Apache-2.0 | MIT |
| Pricing | Open source | Open source |
| Released | 2023-05-31 | 2025-04-28 |
| Paper | Link | Link |
| GitHub | — | GitHub |
| GitHub stars | — | 168 |
| Interface | — | Package |
License and commercial use
| Field | Geneformer | TranscriptFormer |
|---|---|---|
| Code license | Apache-2.0 | MIT |
| Weights license | — | — |
| Commercial use | Permissive (commercial OK) | Permissive (commercial OK) |
| Answer | Yes. The recorded license is permissive, so closed-source commercial use is allowed if you keep the notices. Details | Yes. The recorded license is permissive, so closed-source commercial use is allowed if you keep the notices. Details |
Add more Single-cell & foundation models tools to this comparison →