Citation

Citing scFoundry

If this framework or any of the tools provided here are useful for your research, please cite our work — it helps us a lot.

Siyu Hou, Penghui Yang, Wenjing Ma, Jinxi Xiang, Jade Xiaoqing Wang, Hui Wan, Ying Ma and Xiang Zhou (2026). Accessible and reproducible deployment reveals the practical boundaries of single-cell foundation models. bioRxiv. https://doi.org/10.64898/2026.01.06.698060

Note

The DOI is that of the bioRxiv preprint. The revised manuscript carries this title and author list; it will replace the version currently online within the next one to two weeks, and until then the DOI resolves to the earlier title.

@article{hou2026accessible,
  title = {Accessible and reproducible deployment reveals the practical boundaries of single-cell foundation models},
  author = {Hou, Siyu and Yang, Penghui and Ma, Wenjing and Xiang, Jinxi and Wang, Jade Xiaoqing and Wan, Hui and Ma, Ying and Zhou, Xiang},
  year = {2026},
  publisher = {Cold Spring Harbor Laboratory},
  journal = {bioRxiv},
  doi = {10.64898/2026.01.06.698060},
  url = {https://doi.org/10.64898/2026.01.06.698060}
}

Citing the models you ran

scFoundry is a deployment and evaluation layer. The models themselves are the work of other groups, and using one through this framework does not replace citing it.

Important

Cite the upstream paper for every method whose results you report. The table below links each project so you can pick up its current citation; version numbers are the ones scFoundry pins.

Method	Version	Upstream project
Cell2Sentence (C2S)	v1.2.0	https://github.com/vandijklab/cell2sentence
CELLama	v0.1.0	https://github.com/portrai-io/CELLama
CellFM	commit 5054a2a	https://github.com/biomed-AI/CellFM-torch
CellPLM	v0.1.0	https://github.com/OmicsML/CellPLM
Geneformer	V2-316M	https://huggingface.co/ctheodoris/Geneformer
GenePT	commit 3602699	https://github.com/yiqunchen/GenePT
LangCell	commit 69e41ef	https://github.com/PharMolix/LangCell
Novae	novae-human-0	https://github.com/MICS-Lab/novae
scBERT	v1.0.0	https://github.com/TencentAILabHealthcare/scBERT
scCello	commit 767585b	https://github.com/DeepGraphLearning/scCello
scFoundation	commit 397631c	https://github.com/biomap-research/scFoundation
scGPT	v0.2.4	https://github.com/bowang-lab/scGPT
SCimilarity	v0.4.1, model v1.1	https://genentech.github.io/scimilarity/
scPRINT	2.3.8, checkpoint medium-v1.5	https://github.com/cantinilab/scPRINT
scVI	Census 2024-07-01 (Homo sapiens)	https://virtualcellmodels.cziscience.com/model/scvi-homo-sapiens
UCE	commit 8227a65	https://github.com/snap-stanford/UCE

Citing the integration baselines

The integration methods of embed have their own references:

Method	Upstream project
Harmony	https://github.com/immunogenomics/harmony
Seurat v5 (CCA and RPCA integration)	https://satijalab.org/seurat/
scVI (de novo integration)	https://scvi-tools.org/

Citing the metrics

The evaluation metrics come from the single-cell integration benchmarking literature and are computed with scib-metrics and scikit-learn. If you report metric values, cite scib-metrics alongside the original benchmark papers for the specific metrics you use. Definitions and provenance for each metric are on the evaluation metrics page. The representation-geometry probes are defined in the accompanying paper; the TwoNN intrinsic dimension estimator is due to Facco et al. (2017) and the R_NX criterion to Lee and Verleysen (2009), both cited on the probes page.

Data

The benchmark in the paper is built on Tabula Sapiens data obtained from the CZ CELLxGENE Discover portal. If you reuse the prepared datasets, cite the Tabula Sapiens consortium and CZ CELLxGENE Discover as well — see Benchmark design for the exact collection.