CoChat
Sign Up
Cited by 46Single-cell and spatial transcriptomicsBioinformatics and Genomic NetworksGene Regulatory Network Analysis

Single-cell foundation models: bringing artificial intelligence into cell biology

Open access
Baek, Seungbyn, Song, Kyungwoo, Lee, Insuk·2025·Experimental & Molecular Medicine
DOI 10.1038/s12276-025-01547-5PMC PMC12586647
Open this paper in CoChat — free accountOpen PDF

Abstract

A foundation model, a large-scale deep learning model pretrained on vast datasets, has revolutionized data interpretation through self-supervised learning with capacity for various downstream tasks. Concurrently, single-cell genomics is in urgent need of unified frameworks capable of integrating and comprehensively analyzing the rapidly expanding data repositories. Inspired by advances in foundation models, researchers have extended these techniques to single-cell analysis, giving rise to single-cell foundation models (scFMs). Typically, these models use transformer architectures to incorporate diverse omics data and extract latent patterns at both cell and gene/feature levels for the analysis of cellular heterogeneity and complex regulatory networks. Despite their promise, scFMs face challenges including the nonsequential nature of omics data, inconsistency in data quality and the computational intensity required for training and fine-tuning. Furthermore, interpreting the biological relevance of latent embeddings and model representations remains nontrivial. Here we provide an overview of scFMs, highlighting their key concepts and applications across downstream tasks. We critically assess the current limitations and propose future directions aimed at enhancing the robustness, interpretability and scalability of scFMs. Ultimately, addressing these challenges will be crucial for establishing scFMs as pivotal tools in advancing single-cell genomics and unlocking deeper insights into cellular function and disease mechanisms.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s12276-025-01547-5

Cites

16 of 16 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Recent work suggests that architectural choices, such as generative versus contrastive pretraining objectives, introduce distinct trade-offs between representation flexibility, interpretability, and robustness to experimental variability.

    “Recent work suggests that architectural choices, such as generative versus contrastive pretraining objectives, introduce distinct trade-offs between representation flexibility, interpretability, and robustness to experimental variability [20–23, 36].”

    From The next paradigm in bioinformatics: a review of multi-agent systems and foundational models for end-to-end scientific discovery (2026) · cites this paper as [36]

    The Pretraining strategies section (page unavailable) describes different effects of gene-prediction and contrastive objectives on learned representations, and the paper separately discusses interpretability and batch-related variability. But the Model architecture section says these architectural strengths have not been clearly compared in single-cell foundation models. The paper therefore does not establish the claimed trade-offs between flexibility, interpretability, and robustness to experimental variability. (LLM, conf=0.86)

    Experimental — accuracy not yet measured

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 44

  1. [3]Guo, F. et al. Foundation models in bioinformatics. Natl Sci. Rev.10.1093/nsr/nwaf028 (2025). PMC11900445 40078374 · DOI
  2. [4]Zhou, C. et al. A comprehensive survey on pretrained foundation models: a history from BERT to ChatGPT. Int. J. Mach. Learn. Cybern.10.1007/s13042-024-02443-6 (2024). · DOI
  3. [6]Hollmann, N. et al. Accurate predictions on small data with a tabular foundation model. Nature 637, 319–326 (2025). 39780007 PMC11711098 · DOI
  4. [8]Fu, X. et al. A foundation model of transcription across human cell types. Nature637, 965–973 (2025). PMC11754112 39779852
  5. [9]Chen, X. et al. EpiAgent: foundation model for single-cell epigenomic data. Preprint at bioRxiv10.1101/2024.12.19.629312 (2024).
  6. [10]Wang, C. X. et al. scGPT-spatial: continual pretraining of single-cell foundation model for spatial transcriptomics. Preprint at bioRxiv10.1101/2025.02.05.636714 (2025).
  7. [11]Schaar, A. C. et al. Nicheformer: a foundation model for single-cell and spatial omics. Preprint at bioRxiv10.1101/2024.04.15.589472 (2024). PMC12695652 41168487
  8. [12]Xu, J., Huang, D. S. & Zhang, X. scmFormer integrates large-scale single-cell proteomics and transcriptomics data by multi-task transformer. Adv. Sci. 11, 2307835 (2024). PMC11109621 38483032 · DOI
  9. [13]Yang, F. et al. scBERT as a large-scale pretrained deep language model for cell type annotation of single-cell RNA-seq data. Nat. Mach. Intell. 4, 852–866 (2022). · DOI · Paper page
  10. [15]Zhao, S., Zhang, J. & Nie, Z. Large-scale cell representation learning via divide-and-conquer contrastive learning. Preprint at 10.48550/arXiv.2306.04371 (2023). · DOI
  11. [16]Rosen, Y. et al. Universal cell embeddings: a foundation model for cell biology. Preprint at bioRxiv10.1101/2023.11.28.568918 (2024).
  12. [18]Chen, H. et al. Quantized multi-task learning for context-specific representations of gene network dynamics. Preprint at bioRxiv10.1101/2024.08.16.608180 (2024).
  13. [19]Kalfon, J., Samaran, J., Peyré, G. & Cantini, L. scPRINT: pre-training on 50 million cells allows robust gene network predictions. Nat. Commun. 16, 3607 (2025). 40240364 PMC12003772 · DOI
  14. [20]Program, C. C. S. et al. CZ CELLxGENE Discover: a single-cell data platform for scalable exploration, analysis and modeling of aggregated data. Nucleic Acids Res. 53, D886–D900 (2025). 39607691 PMC11701654 · DOI · Paper page
  15. [22]Franzén, O., Gan, L.-M. & Björkegren, J. L. PanglaoDB: a web server for exploration of mouse and human single-cell RNA sequencing data. Database 2019, baz046 (2019). 30951143 PMC6450036 · DOI · Paper page
  16. [23]Chen, S. et al. hECA: the cell-centric assembly of a cell atlas. iScience 25, 104318 (2022). 35602947 PMC9114628 · DOI
  17. [24]DenAdel, A. et al. Evaluating the role of pre-training dataset size and diversity on single-cell foundation model performance. Preprint at bioRxiv10.1101/2024.12.13.628448 (2024).
  18. [25]Nadig, A. et al. Consequences of training data composition for deep learning models in single-cell biology. Preprint at bioRxiv10.1101/2025.02.19.639127 (2025).
  19. [26]Feng, S., Li, S., Chen, L. & Chen, S. Unveiling potential threats: backdoor attacks in single-cell pre-trained models. Cell Discov. 10, 122 (2024). 39616169 PMC11608368 · DOI
  20. [27]Alsabbagh, A. R. et al. Foundation models meet imbalanced single-cell data when learning cell type annotations. Preprint at bioRxiv10.1101/2023.10.24.563625 (2023).
  21. [28]Zeng, Y. et al. CellFM: a large-scale foundation model pre-trained on transcriptomics of 100 million human cells. Nat. Commun. 16, 1–17 (2025). 40393991 PMC12092794 · DOI
  22. [29]Shen, H. et al. Generative pretraining from large-scale transcriptomes for single-cell deciphering. iScience 26, 106536 (2023). 37187700 PMC10176267 · DOI
  23. [30]Chevalier, A. et al. TEDDY: a family of foundation models for understanding single cell biology. Preprint at 10.48550/arXiv.2503.03485 (2025). · DOI
  24. [31]Fang, C. et al. Cell-Graph Compass: modeling single cells with graph structure foundation model. Preprint at bioRxiv10.1101/2024.06.04.597354 (2024).
  25. [32]Vaswani, A. et al. Attention is all you need. Adv. Neural Inf. Process Syst. 30, 5998–6008 (2017).
  26. [35]Wang, T. et al. What language model architecture and pretraining objective works best for zero-shot generalization?. In Proceedings of the 39th International Conference on Machine Learning, PMLR 162:22964–22984 (2022).
  27. [36]Tay, Y. et al. UL2: unifying language learning paradigms. Preprint at 10.48550/arXiv.2205.05131 (2022). · DOI
  28. [37]Bian, H. et al. General-purpose pre-trained large cellular models for single-cell transcriptomics. Natl Sci. Rev. 11, nwae340 (2024). 39555105 PMC11565237 · DOI
  29. [38]Heimberg, G. et al. A cell atlas foundation model for scalable search of similar human cells. Nature 638, 1085–1094 (2025). 39566551 PMC11864978 · DOI
  30. [39]Hu, E. J. et al. LoRA: low-rank adaptation of large language models. ICLR 1, 3 (2022).
  31. [40]Dettmers, T., Pagnoni, A., Holtzman, A. & Zettlemoyer, L. QLoRA: efficient finetuning of quantized LLMs. Adv. Neural Inf. Process. Syst. 36, 10088–10115 (2023).
  32. [42]Qiu, P. et al. BioLLM: A standardized framework for integrating and benchmarking single-cell foundation models. Patterns 6, 101326 (2025). 40843339 PMC12365531 · DOI
  33. [45]Gong, W., Kwak, I.-Y., Pota, P., Koyano-Nakagawa, N. & Garry, D. J. DrImpute: imputing dropout events in single cell RNA sequencing data. BMC Bioinform. 19, 1–10 (2018). PMC5994079 29884114 · DOI
  34. [46]Sung, E., Cha, J., Baek, S. & Lee, I. Augmenting the human interactome for disease prediction through gene networks inferred from human cell atlas. Anim. Cells Syst. 29, 11–20 (2025). PMC11892045 40066175 · DOI
  35. [48]Wen, H. et al. CellPLM: pre-training of cell language model beyond single cells. Preprint at bioRxiv10.1101/2023.10.03.560734 (2023).
  36. [49]Janssen, P. et al. The effect of background noise and its removal on the analysis of single-cell expression data. Genome Biol. 24, 140 (2023). 37337297 PMC10278251 · DOI
  37. [50]Montserrat-Ayuso, T. & Esteve-Codina, A. High content of nuclei-free low-quality cells in reference single-cell atlases: a call for more stringent quality control using nuclear fraction. BMC Genom. 25, 1124 (2024). PMC11580415 39574015 · DOI
  38. [52]Preissl, S., Gaulton, K. J. & Ren, B. Characterizing cis -regulatory elements using single-cell epigenomics. Nat. Rev. Genet. 24, 21–43 (2023). 35840754 PMC9771884 · DOI · Paper page
  39. [54]Vandereyken, K., Sifrim, A., Thienpont, B. & Voet, T. Methods and applications for single-cell and spatial multi-omics. Nat. Rev. Genet. 24, 494–515 (2023). 36864178 PMC9979144 · DOI · Paper page
  40. [55]Bennett, H. M., Stephenson, W., Rose, C. M. & Darmanis, S. Single-cell proteomics enabled by next-generation sequencing or mass spectrometry. Nat. Methods 20, 363–374 (2023). 36864196 · DOI · Paper page
  41. [56]Baysoy, A., Bai, Z., Satija, R. & Fan, R. The technological landscape and applications of single-cell multi-omics. Nat. Rev. Mol. Cell Biol. 24, 695–713 (2023). 37280296 PMC10242609 · DOI · Paper page
  42. [57]Minaee, S. et al. Large language models: a survey. Preprint at 10.48550/arXiv.2402.06196 (2024). · DOI · Paper page
  43. [58]Das, B. C., Amini, M. H. & Wu, Y. Security and privacy challenges of large language models: a survey. ACM Comput. Surv. 57, 1–39 (2025).
  44. [59]Ding, J. et al. Toward a privacy-preserving predictive foundation model of single-cell transcriptomics with federated learning and tabular modeling. Preprint at bioRxiv10.1101/2025.01.06.631427 (2025).

Cited by

Sorted by citations

Authors
Baek, S., Song, K., & Lee, I.
Affiliations
Yonsei University (KR)
Corresponding author
Lee, Insuk
ORCID
Venue
Experimental & Molecular Medicine
Publisher
Springer Science and Business Media LLC
Published
2025-10-01
Volume
57
Issue
10
Pages
2169-2181
ISSN
  • 1226-3613 (ISSN-L)
  • 2092-6413
DOI
10.1038/s12276-025-01547-5
PMC
PMC12586647
OpenAlex
W4414734186
Cited by
46
References
16
Citations by year
2026: 44 · 2025: 2
FWCI
4.07
Type
journal article
Language
en
Open access
Licence
Locations
  • Experimental & Molecular Medicine (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
National Research Foundation of Korea (2022M3A9F3016364, 2022R1A2C1092062)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
Keywords
single-cell foundation models, single-cell genomics, transformer models, self-supervised learning, multi-omics integration, cellular heterogeneity, gene regulatory networks, latent space embedding, model interpretability, computational scalability
MeSH terms
Deep Learning, Animals, Artificial Intelligence, Cell Biology, Humans, Models, Biological, Computational Biology, Genomics, Single-Cell Analysis

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review