scGPT Evaluation Suite

Evaluation shipped with scGPT foundation model (cell type annotation, GRN, perturbation).

Composite score: 73.7

Rubric (1–5 per criterion)

rigor

3/5

coverage

3/5

maintenance

4/5

adoption

5/5

quality

3/5

accessibility

5/5

industry_relevance

3/5

Metadata

Stages

Virtual Cell

Modalities

cross-modality

Task types

cell-type-annotationperturbation-prediction

License

MIT

First release

2023-05

Last updated

2025-01

Flags

self_referential

Size & scope

cells: 33000000
tasks: 5

Primary paper

Title

scGPT: toward building a foundation model for single-cell multi-omics using generative AI

Authors

Cui H, Wang C, Maan H, Pang K, Luo F, Wang B

Year

2024

DOI / arXiv

10.1038/s41592-024-02201-0

Citations

520

Hosted by (initiatives)

not hosted by any tracked initiative

Experts (primary authors / maintainers)

Bo Wang

Groups (host labs / companies / consortia)

Bo Wang Lab (Vector/UofT)

Related benchmarks

Open Problems: Perturbation Prediction

Notes (honest caveats)

Evaluation dominated by authors' own model — flagged self-referential. Pair with OpenProblems for fair comparison.