Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

SToFM: a Multi-scale Foundation Model for Spatial Transcriptomics

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read SToFM claims a multi-scale spatial transcriptomics representation that fuses gene expression, local cell-cell interactions, and tissue morphology, and reports 11.6–14.34% F1 gains over the best baselines in cross-slice tasks.

desk verdict A serious multi-scale ST foundation model with a large new corpus, but the headline gains rest on a train/test split the paper does not yet let anyone verify; worth peer review. read the letter →

arxiv 2507.11588 v2 pith:F3V2EGEI submitted 2025-07-15 q-bio.GN cs.AIcs.LG

Suyuan Zhao , Yizhen Luo , Ganbo Yang , Yan Zhong , Hao Zhou , Zaiqing Nie This is my paper · ORCID
classification q-bio.GNcs.AIcs.LG
keywords spatialtranscriptomicsfoundationmodelmulti-scalerepresentationlearningSE(2)Transformervirtualcellscelltypeannotationtissuesemanticsegmentationpretrainingcorpus
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

SToFM is a pretrained model for spatial transcriptomics whose central claim is that cell representations should integrate three scales at once: each cell's gene expression, its local neighborhood of nearby cells, and the global tissue region it sits in. The paper argues that existing single-cell and spatial foundation models capture at most expression plus coordinates, and that missing macro- and micro-scale structure is why they transfer poorly across slices and technologies. To test this, the authors pretrain on SToCorpus-88M, roughly 88 million cells from about 2,000 high-resolution slices spanning six sequencing platforms, and then evaluate on tissue-region segmentation, cell type annotation, clustering, deconvolution, and imputation. SToFM reports the best results on all of these tasks, with the largest margins in cross-slice transfer: 11.6% higher macro F1 than the best baseline on cross-slice tissue-region segmentation and 14.34% higher macro F1 on cell type annotation. If the results hold, the model would provide biologists with a single pretrained representation that follows consistent cell-type and tissue-region semantics across different experiments and technologies.

What carries the argument

The load-bearing mechanism is the virtual-cell sub-slice: a tissue slice is divided by spatial location into sub-slices of roughly 1,000 cells, while Leiden clustering over a blend of cell embeddings and positions produces cluster centroids, called virtual cells, that are inserted into every sub-slice. Each virtual cell's embedding and position are the averages of its cluster, so a few dozen virtual cells summarize the slice's morphology and functional zones. The SE(2) Transformer then takes the real and virtual cells of a sub-slice, converts pairwise Euclidean distances into attention bias through a Gaussian module, and is pretrained with masked cell modeling and pairwise distance recovery. This design makes the cell representations invariant to 2D translation and rotation while letting local cell-cell interactions and global tissue structure jointly shape each cell's embedding.

What would settle it

Compare the released SToCorpus-88M against the exact downstream test slices—CS12-13E2S4 for EmbryoCross, DLPFC slice 151673, the two mouse-brain annotation slices, liver slices D17_FR4 and D0_DY2, and skin Sample1—and check whether any of their cells appear in the pretraining corpus; if they do, the cross-slice improvements would be explained by pretraining on the test data.

Watch

Extended reading notes

Core claim

The paper's discovery is that a spatial transcriptomics foundation model can learn transferable cell representations by compressing each tissue slice into sub-slices that carry gene-scale embeddings, micro-scale local neighborhoods, and macro-scale virtual cells, and then training an SE(2) Transformer on those sub-slices. The macro-scale is encoded by clustering all cells in a slice and inserting each cluster's averaged embedding and position as a virtual cell into every sub-slice, so the model can perceive global tissue morphology while attending to a local group of about 1,000 cells. The gene scale comes from a cell encoder that is first adapted to spatial transcriptomics data; the micro scale comes from transformer layers whose attention is biased by pairwise distances. Pretraining uses two objectives: masked cell modeling, which reconstructs randomly masked cell embeddings from surrounding context, and pairwise distance recovery, which removes Gaussian noise added to cell coordinates. The result is a representation that the paper shows transfers across slices and technologies, outperforming the best baselines in cross-slice semantic segmentation, cell type annotation, clustering, deconvolution, and imputation.

Load-bearing premise

The load-bearing premise is that the slices used in downstream evaluation were actually removed from SToCorpus-88M before pretraining; the paper states this removal but provides no auditable list, so if any evaluated slice leaked into the corpus, the reported transfer gains could come from memorizing test data rather than from multi-scale representation learning.

Editorial extensions

If this is right

  • A classification head trained on some slices can label a held-out slice from the same tissue, and SToFM's advantage over baselines is largest in this cross-slice setting, indicating that the representations carry slice-invariant tissue structure.
  • The model transfers to a sequencing technology absent from pretraining: on 10x Visium DLPFC data, SToFM outperforms baselines that had also never seen Visium, suggesting adaptation to new platforms.
  • The same pretrained embeddings support regression tasks beyond classification, with gains in cross-slice deconvolution and imputation.
  • Scaling the pretraining corpus from 12.5% to 100% of SToCorpus-88M improves downstream F1 in the paper's ablations, tying transferability to corpus scale.
  • Removing any one of the three scales—domain-adapted gene embeddings, micro-scale sub-slice interactions, or macro-scale virtual cells—degrades performance, supporting the multi-scale design over single-scale alternatives.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the corpus-exclusion check passes, the virtual-cell approach is a general recipe for very large 2D point clouds: global context can be injected as a handful of averaged prototypes instead of global attention, which should extend to spatial proteomics, spatial chromatin accessibility, or large histopathology image patch graphs.
  • A sharper test of multi-scale necessity would be a same-size comparison: hold the cell encoder and pretraining budget fixed and compare SToFM against an equally large architecture that only sees expression plus raw coordinates; the paper's ablations approximate this but do not match parameter counts across baselines.
  • Because all main evaluations come from tissues and technologies present in the corpus except Visium, a natural next experiment is to hold out an entire technology and tissue type and measure the drop in cross-technology transfer.
  • If the released corpus keeps slice identifiers, one could test how strongly macro-scale information drives transfer by varying the Leiden clustering resolution across many tissues and correlating downstream F1 with the number of virtual cells.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes SToFM, a multi-scale foundation model for spatial transcriptomics that integrates gene-scale expression embeddings (via a domain-adapted Geneformer encoder), micro-scale spatial context (via an SE(2) Transformer operating on sub-slices of roughly 1,000 cells), and macro-scale tissue structure (via Leiden-derived virtual cells injected into each sub-slice). The model is pretrained with two self-supervised objectives, masked cell modeling and pairwise distance recovery, on a new corpus called SToCorpus-88M containing about 88 million cells from six spatial technologies. The authors report consistent gains over Geneformer, scGPT, Nicheformer, and CellPLM across tissue-region segmentation, cell-type annotation, zero-shot clustering, deconvolution, and imputation, with headline improvements of 11.6% macro-F1 on cross-slice embryo segmentation and 14.34% macro-F1 on cell-type annotation. The paper also includes ablations supporting the contribution of domain adaptation, virtual cells, the micro-scale encoder, and the spatial distance matrix, plus additional experiments on full-parameter tuning, data volume, and hyperparameter sensitivity.

Significance. If the empirical claims hold, this is a substantial step for spatial transcriptomics foundation models: it is, to my knowledge, the first ST foundation model that explicitly represents macro-scale tissue morphology through virtual cells while jointly modeling micro-scale cell-cell interactions and gene-scale expression, and SToCorpus-88M would be the largest public high-resolution ST pretraining corpus to date. The self-supervised objectives are natural and the evaluation spans several practically important tasks. The paper also provides useful ablation evidence and promises code and data release, which strengthens reproducibility if the release is complete. The principal limitation is that the headline comparison rests on an unverifiable train/test separation between SToCorpus-88M and the downstream validation slices, and on hyperparameter choices made using the same downstream tasks that are later reported as the main results. These issues are fixable with additional disclosure and validation, but they currently prevent me from treating the quantitative claims as established.

major comments (4)
  1. [§4.1, Appendix D.2] The separation between SToCorpus-88M and the downstream evaluation slices is asserted but not auditable. Section 4.1 states that pretraining is performed after removing data used for downstream tasks, and Appendix D.2 repeats that individual downstream slices are excluded, but no slice-level manifest, overlap check, or removal script is provided. This is a material risk because the downstream datasets are drawn from the same public repositories used to construct the corpus (notably STOmicsDB STDS0000139 and STDS0000239, HESTA, SODB, Vizgen, 10x Xenium, and a non-public SeekSpace slice). A single overlapping slice could plausibly account for a large part of the reported F1 gains in the abstract and Section 4. Please release the full corpus manifest with unique slice identifiers, the exclusion lists for all downstream evaluation slices, the deduplication procedure, and any coordinate- or identifier-based overlap checks.
  2. [Appendix B.2, Tables A.2, A.6, A.7] Several central hyperparameters were selected on the same downstream tasks used to report the headline results. The sub-slice scale and Leiden resolution (Table A.2), the mixing weight α in Algorithm 1 (Table A.6), and the loss ratio γ (Table A.7) are all evaluated on Embryo2 and EmbryoCross macro-F1, and the same tasks appear in the main results (Tables 1 and 5) and in the abstract's headline gain. This selective reporting risks inflating SToFM's apparent advantage over baselines whose hyperparameters were not similarly tuned on the evaluation tasks. Please either freeze all method hyperparameters before inspecting downstream labels, or report a nested validation procedure (for example, tuning on a hold-out subset and testing on a disjoint subset with rotations), or provide a sensitivity analysis showing that every reasonable hyperparameter choice preserves the reported comparative conclusions.
  3. [Appendix C.2, Tables 1–5] The main quantitative claims are reported without any measure of variability, although Appendix C.2 says that tasks with inherent randomness were run three times and averaged. Tables 1–5 contain only point estimates, with no standard deviations, confidence intervals, or significance tests. This matters because several comparisons are small in magnitude or slice-specific: for example, DLPFC4 macro-F1 is 0.6274 for SToFM versus 0.5918 for CellPLM (Table 1), and the LiverCross MAE difference is 0.0430 versus 0.0441 (Table 3). Please report the three individual runs or summary statistics, and state explicitly whether each reported improvement is consistent across all runs.
  4. [§4.1, Tables 1–4] The baseline comparison conflates architecture with pretraining data scale and domain adaptation. Geneformer, scGPT, Nicheformer, and CellPLM are used as released, with no continual pretraining on SToCorpus-88M, while SToFM combines Geneformer initialization, an additional domain-adaptation phase, and multi-scale pretraining on 88 million cells. The observed gains could therefore be due partly to the larger and task-aligned pretraining corpus rather than to the proposed multi-scale architecture itself. Please either continue-pretrain the baseline models on the same SToCorpus-88M data under the same exclusion rules, or explicitly discuss this confound and provide an additional controlled comparison where all models share the same cell-encoder initialization and pretraining data.
minor comments (4)
  1. [Title page] The affiliation for Tsinghua University contains the typo 'Tecnology' and should read 'Technology'.
  2. [Table A.7] The caption of Table A.7 says 'Ablation study of α in Algorithm 1', but the table actually varies the loss ratio γ; the caption should be corrected to avoid confusion with Table A.6.
  3. [Section 3.3] The MCM and PDR loss equations write the squared L2 norm inside an outer square, which makes the effective exponent ambiguous; please use a single consistent notation for the squared-loss objective.
  4. [Appendix D.2] The SeekSpace slice used for cell-type annotation is described as non-public, while the paper states that SToCorpus-88M and code will be released; please state the exact release timeline and conditions for the SeekSpace data, since the cell-type annotation result depends on it.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: SToFM's pretraining objectives are self-supervised and downstream claims are external-benchmark comparisons.

full rationale

The claimed derivation chain is not circular. SToFM is trained on SToCorpus-88M with two self-supervised objectives (masked cell modeling and pairwise distance recovery) that do not use downstream labels: MCM reconstructs masked cell embeddings from unmasked neighbors, and PDR reconstructs an unperturbed distance matrix from noise-augmented coordinates. Neither objective defines the downstream F1 metric in terms of a fitted parameter. Downstream tasks (embryo/DLPFC segmentation, cell-type annotation, clustering, deconvolution, imputation) are evaluated against external baselines (Geneformer, scGPT, Nicheformer, CellPLM) on datasets external to pretraining, and the paper states that downstream slices were removed before pretraining. The same-author citations (e.g., Zhao et al. for a contrastive objective) are not load-bearing: they suggest a training component but do not constitute a uniqueness theorem or forced choice. The main verifiability risk is that the removal of downstream slices from SToCorpus-88M is asserted without a public manifest; that is a data-leakage and auditability concern, not a circularity step, because it would not make the reported result equal to its input by construction. No equation or fitted quantity in the paper reduces a prediction to the target it claims to predict.

Assumptions & free parameters 8 free parameters · 5 assumptions · 1 invented entities

The central claim rests on a standard Transformer architecture with distance-based attention, a domain-adapted Geneformer encoder, and a large pretraining corpus. The model's design introduces several free hyperparameters, several of which were selected using the same downstream benchmarks that define the main results. Virtual cells are the only genuinely new entity, and they lack independent evidence beyond the paper's own ablations.

free parameters (8)
  • Leiden clustering resolution = 1.0
    Selected via ablation on the downstream tasks Embryo2 and EmbryoCross (Table A.2); the final value is optimal on the very benchmarks reported in Table 1.
  • alpha (α) in Algorithm 1 = 0.8
    Selected via ablation on Embryo2 and EmbryoCross (Table A.6), i.e., tuned on the reported benchmarks.
  • Sub-slice scale (cells per sub-slice) = 1000
    Chosen via ablation (Table A.2) as the best trade-off between computational cost and performance.
  • Sample number for second-time cell encoding = 12
    Set to the Geneformer training batch size; robustness tested on 1/8 of the corpus (Table A.4).
  • Loss ratio gamma (γ) = 0.5
    Chosen from ablation on 1/8 of the corpus (Table A.7); 1:1 ratio, no evidence of strong sensitivity.
  • MCM masking ratio = 0.10
    Fixed in Section 3.3; no ablation reported.
  • PDR perturbation ratio = 0.10
    Fixed in Section 3.3; no ablation reported.
  • PCA dimension for clustering input = 2
    Algorithm 1 reduces F to 2D via PCA before combining with positions; fixed choice.
assumptions (5)
  • domain assumption A spatial transcriptomics slice can be represented as a 2D point cloud with a gene expression vector per point.
    Section 3.1 defines S0 = (E, P); this representation ignores tissue geometry beyond coordinates, cell segmentation errors, and any spatial correlation structure not captured by pairwise distances.
  • domain assumption Geneformer's rank-based tokenization and weights transfer to spatial transcriptomics after one epoch of domain adaptation.
    Section 3.1 and Appendix A.1 assume the scRNA-seq pretrained encoder provides a good initialization for ST gene expression despite dropout zeros and limited gene coverage.
  • ad hoc to paper Leiden clustering on alpha*Normalize(PCA2(F)) + (1-alpha)*Normalize(P) yields clusters that faithfully encode macro-scale tissue structure.
    Algorithm 1 assumes averaged cluster embeddings (virtual cells) are a valid compression of tissue morphology; this is the paper's own design choice, tested only by internal ablations.
  • domain assumption Cell-cell interactions in ST are sufficiently local that sub-slicing into ~1000-cell windows preserves the micro-scale information needed for downstream tasks.
    Section 3.1 uses spatial sub-slices to limit the receptive field; this assumption is supported by a cited reference (Zormpas et al., 2023) but is not directly verified on the benchmarks.
  • domain assumption SE(2) invariance is appropriate for the learned representations; orientation of the tissue slice carries no useful signal.
    Section 3.2 uses only the distance matrix as spatial input to the SE(2) Transformer, discarding absolute coordinates and orientation.
invented entities (1)
  • Virtual cells
    purpose: Aggregate clusters of real cells (by embedding and position averages) to inject macro-scale tissue structure into every sub-slice.
    Virtual cells are an internal construct of SToFM; their utility is supported only by the model's own ablation studies (Table 5, A.2), with no external or measured evidence of their existence or necessity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SToFM: a Multi-scale Foundation Model for Spatial Transcriptomics." pith.science (2026). https://pith.science/paper/F3V2EGEI

@misc{pith2026250711588,
  author       = {Pith},
  title        = {Pith review of: SToFM: a Multi-scale Foundation Model for Spatial Transcriptomics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F3V2EGEI}},
  note         = {Machine review of arXiv:2507.11588}
}
read the original abstract

Spatial Transcriptomics (ST) technologies provide biologists with rich insights into single-cell biology by preserving spatial context of cells. Building foundational models for ST can significantly enhance the analysis of vast and complex data sources, unlocking new perspectives on the intricacies of biological tissues. However, modeling ST data is inherently challenging due to the need to extract multi-scale information from tissue slices containing vast numbers of cells. This process requires integrating macro-scale tissue morphology, micro-scale cellular microenvironment, and gene-scale gene expression profile. To address this challenge, we propose SToFM, a multi-scale Spatial Transcriptomics Foundation Model. SToFM first performs multi-scale information extraction on each ST slice, to construct a set of ST sub-slices that aggregate macro-, micro- and gene-scale information. Then an SE(2) Transformer is used to obtain high-quality cell representations from the sub-slices. Additionally, we construct \textbf{SToCorpus-88M}, the largest high-resolution spatial transcriptomics corpus for pretraining. SToFM achieves outstanding performance on a variety of downstream tasks, such as tissue region semantic segmentation and cell type annotation, demonstrating its comprehensive understanding of ST data through capturing and integrating multi-scale information.

Figures

Figures reproduced from arXiv: 2507.11588 by the authors.

Figure 1
Figure 1. ST data contain biological information from multiple scales. (a). An example of a visualized ST slice of a mouse brain. (b). A microscopic sub-slice of the brain slice. (c). Gene expression profile of a single cell in the brain slice. cells, providing a digital representation of cells (Ziegenhain et al., 2017). Machine learning approaches have achieved great success in analyzing scRNA-seq data and enhancing the unde… view at source ↗
Figure 2
Figure 2. An overview of SToFM architecture. (a). Multi-scale processing integrates information on macro, micro, and gene scales into sub-slices. At gene scale, we extract the representations of each cell using a cell encoder that has undergone domain adaptation. At micro scale, we divide the ST slice into several sub-slices. At macro scale, we cluster cells based on spatial and gene expression data, construct virtual cells f… view at source ↗
Figure 3
Figure 3. UMAP visualization of cell representations on a mouse brain slice, colored by cell types. ARI and NMI scores were calculated with Leiden clustering results and cell type labels [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Spatial Transcriptomics as Images for Large-Scale Pretraining

    cs.CV 2026-03 conditional novelty 5.0 of 10

    Cropping ST slides into fixed multi-channel gene patches preserves local spatial context, multiplies training samples, and beats spot- and slice-based pretraining on domain detection.

Reference graph

Works this paper leans on

80 extracted references · 71 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    10xgenomics datasets

    10xGENOMICS. 10xgenomics datasets. https://www.10xgenomics.com/datasets, 2025. Accessed: 2025-01-20

    work page 2025

  3. [3]

    Abdelaal, T., Mourragui, S., Mahfouz, A., and Reinders, M. J. Spage: spatial gene enhancement using scrna-seq. Nucleic acids research, 48 0 (18): 0 e107--e107, 2020

    work page 2020

  4. [4]

    J., Bambrick, J., et al

    Abramson, J., Adler, J., Dunger, J., Evans, R., Green, T., Pritzel, A., Ronneberger, O., Willmore, L., Ballard, A. J., Bambrick, J., et al. Accurate structure prediction of biomolecular interactions with alphafold 3. Nature, pp.\ 1--3, 2024

    work page 2024

  5. [5]

    H., Anderson, C

    Adelson, E. H., Anderson, C. H., Bergen, J. R., Burt, P. J., and Ogden, J. M. Pyramid methods in image processing. RCA engineer, 29 0 (6): 0 33--41, 1984

    work page 1984

  6. [6]

    E., Blosser, T

    Allen, W. E., Blosser, T. R., Sullivan, Z. A., Dulac, C., and Zhuang, X. Molecular and spatial signatures of mouse brain aging at single-cell resolution. Cell, 186 0 (1): 0 194--208, 2023

    work page 2023

  7. [7]

    and Fan, J

    Atta, L. and Fan, J. Computational challenges and opportunities in spatially resolved transcriptomic data analysis. Nature Communications, 12 0 (1): 0 5283, 2021

    work page 2021

  8. [8]

    and Pir, P

    Av s ar, G. and Pir, P. A comparative performance evaluation of imputation methods in spatially resolved transcriptomics data. Molecular Omics, 19 0 (2): 0 162--173, 2023

    work page 2023

Show all 80 references
  1. [9]

    R., Segerstolpe, ., Zhang, M., et al

    Biancalani, T., Scalia, G., Buffoni, L., Avasthi, R., Lu, Z., Sanger, A., Tokcan, N., Vanderburg, C. R., Segerstolpe, ., Zhang, M., et al. Deep learning and alignment of spatially resolved single-cell transcriptomes with tangram. Nature methods, 18 0 (11): 0 1352--1362, 2021

    2021

  2. [10]

    S.-C., Abdulla, S., Aevermann, B., Assis, P., Badajoz, S., Bell, S

    Biology, C. S.-C., Abdulla, S., Aevermann, B., Assis, P., Badajoz, S., Bell, S. M., Bezzi, E., Cakir, B., Chaffer, J., Chambers, S., et al. Cz cellxgene discover: A single-cell data platform for scalable exploration, analysis and modeling of aggregated data. bioRxiv, pp.\ 2023...

    2023

  3. [11]

    Bordukova, M., Makarov, N., Rodriguez-Esteban, R., Schmich, F., and Menden, M. P. Generative artificial intelligence empowers digital twins in drug discovery and clinical trials. Expert Opinion on Drug Discovery, 19 0 (1): 0 33--42, 2024

    2024

  4. [12]

    Gcnet: Non-local networks meet squeeze-excitation networks and beyond

    Cao, Y., Xu, J., Lin, S., Wei, F., and Hu, H. Gcnet: Non-local networks meet squeeze-excitation networks and beyond. In Proceedings of the IEEE/CVF international conference on computer vision workshops, pp.\ 0--0, 2019

    2019

  5. [13]

    S., Rowsell, C., Plataniotis, K

    Chan, L., Hosseini, M. S., Rowsell, C., Plataniotis, K. N., and Damaskinos, S. Histosegnet: Semantic segmentation of histological tissue type in whole slide images. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 10662--10671, 2019

    2019

  6. [14]

    Spatiotemporal transcriptomic atlas of mouse organogenesis using dna nanoball-patterned arrays

    Chen, A., Liao, S., Cheng, M., Ma, K., Wu, L., Lai, Y., Qiu, X., Yang, J., Xu, J., Hao, S., et al. Spatiotemporal transcriptomic atlas of mouse organogenesis using dna nanoball-patterned arrays. Cell, 185 0 (10): 0 1777--1792, 2022

    2022

  7. [15]

    Stimage-1k4m: A histopathology image-gene expression dataset for spatial transcriptomics

    Chen, J., Zhou, M., Wu, W., Zhang, J., Li, Y., and Li, D. Stimage-1k4m: A histopathology image-gene expression dataset for spatial transcriptomics. ArXiv, pp.\ arXiv--2406, 2024

    2024

  8. [16]

    H., Boettiger, A

    Chen, K. H., Boettiger, A. N., Moffitt, J. R., Wang, S., and Zhuang, X. Spatially resolved, highly multiplexed rna profiling in single cells. Science, 348 0 (6233): 0 aaa6090, 2015

    2015

  9. [17]

    A cellular resolution spatial transcriptomic landscape of the medial structures in postnatal mouse brain

    Cheng, M., Wu, L., Han, L., Huang, X., Lai, Y., Xu, J., Wang, S., Li, M., Zheng, H., Feng, W., et al. A cellular resolution spatial transcriptomic landscape of the medial structures in postnatal mouse brain. Frontiers in Cell and Developmental Biology, 10: 0 878346, 2022

    2022

  10. [18]

    Cellama: Foundation model for single cell and spatial transcriptomics by cell embedding leveraging language model abilities

    Choi, H., Park, J., Kim, S., Kim, J., Lee, D., Bae, S., Shin, H., and Lee, D. Cellama: Foundation model for single cell and spatial transcriptomics by cell embedding leveraging language model abilities. bioRxiv, pp.\ 2024--05, 2024

    2024

  11. [19]

    X., Maan, H., Pang, K., Luo, F., and Wang, B

    Cui, H., Wang, C. X., Maan, H., Pang, K., Luo, F., and Wang, B. scgpt: Towards building a foundation model for single-cell multi-omics using generative ai. bioRxiv, 2023. URL https://api.semanticscholar.org/CorpusID:258464426

    2023

  12. [20]

    Cross-tissue immune cell analysis reveals tissue-specific features in humans

    Dom \' nguez Conde, C., Xu, C., Jarvis, L., Rainbow, D., Wells, S., Gomes, T., Howlett, S., Suchanek, O., Polanski, K., King, H., et al. Cross-tissue immune cell analysis reveals tissue-specific features in humans. Science, 376 0 (6594): 0 eabl5197, 2022

    2022

  13. [21]

    and Zhang, S

    Dong, K. and Zhang, S. Deciphering spatial domains from spatially resolved transcriptomics with an adaptive graph attention auto-encoder. Nature communications, 13 0 (1): 0 1739, 2022

    2022

  14. [22]

    Advances in spatial transcriptomics and related data analysis strategies

    Du, J., Yang, Y.-C., An, Z.-J., Zhang, M.-H., Fu, X.-H., Huang, Z.-F., Yuan, Y., and Hou, J. Advances in spatial transcriptomics and related data analysis strategies. Journal of Translational Medicine, 21 0 (1): 0 330, 2023

    2023

  15. [23]

    P., Collin, A., Oag, S., Volckaert, T., Borde, A., et al

    Franz \'e n, L., Olsson Lindvall, M., H \"u hn, M., Ptasinski, V., Setyo, L., Keith, B. P., Collin, A., Oag, S., Volckaert, T., Borde, A., et al. Mapping spatially resolved transcriptomes in human and mouse pulmonary fibrosis. Nature Genetics, 56 0 (8): 0 1725--1736, 2024

    2024

  16. [24]

    S im CSE : Simple contrastive learning of sentence embeddings

    Gao, T., Yao, X., and Chen, D. S im CSE : Simple contrastive learning of sentence embeddings. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp.\ 6894--6910, Online and Punta Cana, Dominican Republic, November 2021. Association for C...

    2021 doi

  17. [25]

    xtrimogene: An efficient and scalable representation learner for single-cell rna-seq data

    Gong, J., Hao, M., Cheng, X., Zeng, X., Liu, C., Ma, J., Zhang, X., Wang, T., and Song, L. xtrimogene: An efficient and scalable representation learner for single-cell rna-seq data. arXiv preprint arXiv:2311.15156, 2023

    2023 arXiv

  18. [26]

    T., and Zhang, X

    Hao, M., Gong, J., Zeng, X., Liu, C., Guo, Y., Cheng, X., Wang, T., Ma, J., Song, L. T., and Zhang, X. Large scale foundation model on single-cell transcriptomics. bioRxiv, 2023. URL https://api.semanticscholar.org/CorpusID:259025739

    2023

  19. [27]

    W., Amode, M

    Harrison, P. W., Amode, M. R., Austine-Orimoloye, O., Azov, A. G., Barba, M., Barnes, I., Becker, A., Bennett, R., Berry, A., Bhai, J., et al. Ensembl 2024. Nucleic acids research, 52 0 (D1): 0 D891--D899, 2024

    2024

  20. [28]

    V., Sadick, J

    Hasel, P., Rose, I. V., Sadick, J. S., Kim, R. D., and Liddelow, S. A. Neuroinflammatory astrocyte subtypes in the mouse brain. Nature neuroscience, 24 0 (10): 0 1475--1487, 2021

    2021

  21. [29]

    Detisseg: A dual-encoder network for tissue semantic segmentation of histopathology image

    He, P., Qu, A., Xiao, S., and Ding, M. Detisseg: A dual-encoder network for tissue semantic segmentation of histopathology image. Biomedical Signal Processing and Control, 87: 0 105544, 2024

    2024

  22. [30]

    A., Buhr, D

    He, S., Bhatt, R., Brown, C., Brown, E. A., Buhr, D. L., Chantranuvatana, K., Danaher, P., Dunaway, D., Garrison, R. G., Geiss, G., et al. High-plex imaging of rna and proteins at subcellular resolution in fixed tissue by spatial molecular imaging. Nature Biotechnology, 40 0 (...

    2022

  23. [31]

    Hesta: Human embryogenesis spatiotemporal transcriptomic atlas

    HESTA. Hesta: Human embryogenesis spatiotemporal transcriptomic atlas. db.cngb.org/stomics/hesta/brain/, 2025. Accessed: 2025-01-20

    2025

  24. [32]

    Y., Almagro P \'e rez, C., Wagner, S., Vaidya, A., Chen, R., Williamson, D., Kim, A., et al

    Jaume, G., Doucet, P., Song, A., Lu, M. Y., Almagro P \'e rez, C., Wagner, S., Vaidya, A., Chen, R., Williamson, D., Kim, A., et al. Hest-1k: A dataset for spatial transcriptomics and histology image analysis. Advances in Neural Information Processing Systems, 37: 0 53798--53833, 2024

    2024

  25. [33]

    xmint: A multimodal integration transformer for xenium gene imputation

    Jiang, X., Xie, Y., and Xie, J. xmint: A multimodal integration transformer for xenium gene imputation. In ICML 2024 Workshop on Efficient and Accessible Foundation Models for Biological Discovery, 2024

    2024

  26. [34]

    Advances in spatial transcriptomics and its applications in cancer research

    Jin, Y., Zuo, Y., Li, G., Liu, W., Pan, Y., Fan, T., Fu, X., Yao, X., and Peng, Y. Advances in spatial transcriptomics and its applications in cancer research. Molecular Cancer, 23 0 (1): 0 129, 2024

    2024

  27. [35]

    Highly accurate protein structure prediction with alphafold

    Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Z \' dek, A., Potapenko, A., et al. Highly accurate protein structure prediction with alphafold. nature, 596 0 (7873): 0 583--589, 2021

    2021

  28. [36]

    and Kim, T.-K

    Jung, N. and Kim, T.-K. Spatial transcriptomics in neuroscience. Experimental & Molecular Medicine, 55 0 (10): 0 2105--2115, 2023

    2023

  29. [37]

    A., L \'e vy, S., Pallikkavaliyaveetil, N., Zhang, D., Chen, X., Ghadermarzi, S., Wu, R., Zheng, Z., Vrkic, I., et al

    Levine, D., Rizvi, S. A., L \'e vy, S., Pallikkavaliyaveetil, N., Zhang, D., Chen, X., Ghadermarzi, S., Wu, R., Zheng, Z., Vrkic, I., et al. Cell2sentence: teaching large language models the language of biology. BioRxiv, pp.\ 2023--09, 2023

    2023

  30. [38]

    A comprehensive benchmarking with practical guidelines for cellular deconvolution of spatial transcriptomics

    Li, H., Zhou, J., Li, Z., Chen, S., Liao, X., Zhang, B., Zhang, R., Wang, Y., Sun, S., and Gao, X. A comprehensive benchmarking with practical guidelines for cellular deconvolution of spatial transcriptomics. Nature Communications, 14 0 (1): 0 1548, 2023

    2023

  31. [39]

    St-align: A multimodal foundation model for image-gene alignment in spatial transcriptomics

    Lin, Y., Luo, L., Chen, Y., Zhang, X., Wang, Z., Yang, W., Tong, M., and Yu, R. St-align: A multimodal foundation model for image-gene alignment in spatial transcriptomics. arXiv preprint arXiv:2411.16793, 2024

    2024 arXiv

  32. [40]

    Evaluating the utilities of foundation models in single-cell data analysis

    Liu, T., Li, K., Wang, Y., Li, H., and Zhao, H. Evaluating the utilities of foundation models in single-cell data analysis. bioRxiv, pp.\ 2023--09, 2023

    2023

  33. [41]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., and Guo, B. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 10012--10022, 2021

    2021

  34. [42]

    Method of the year: spatially resolved transcriptomics

    Marx, V. Method of the year: spatially resolved transcriptomics. Nature methods, 18 0 (1): 0 9--14, 2021

    2021

  35. [43]

    R., Collado-Torres, L., Weber, L

    Maynard, K. R., Collado-Torres, L., Weber, L. M., Uytingco, C., Barry, B. K., Williams, S. R., Catallini, J. L., Tran, M. N., Besich, Z., Tippani, M., et al. Transcriptome-scale spatial gene expression in the human dorsolateral prefrontal cortex. Nature neuroscience, 24 0 (3):...

    2021

  36. [44]

    Cosmx™ smi datasets

    nanoString. Cosmx™ smi datasets. https://nanostring.com/products/cosmx-spatial-molecular-imager/ffpe-dataset/, 2025. Accessed: 2025-01-20

    2025

  37. [45]

    Spatiotemporal transcriptome atlas of human embryos after gastrulation

    Pan, J., Li, Y., Lin, Z., Lan, Q., Chen, H., Zhai, M., Sui, S., Zhang, G., Cheng, Y., Tang, Y., et al. Spatiotemporal transcriptome atlas of human embryos after gastrulation. bioRxiv, pp.\ 2023--04, 2023

    2023

  38. [46]

    C., Tejada-Lapuerta, A., Palla, G., Gutgesell, R., Halle, L., Minaeva, M., Vornholz, L., Dony, L., Drummer, F., Bahrami, M., et al

    Schaar, A. C., Tejada-Lapuerta, A., Palla, G., Gutgesell, R., Halle, L., Minaeva, M., Vornholz, L., Dony, L., Drummer, F., Bahrami, M., et al. Nicheformer: a foundation model for single-cell and spatial omics. bioRxiv, pp.\ 2024--04, 2024

    2024

  39. [47]

    Seekspace single-cell transcriptomics technology

    SeekGene. Seekspace single-cell transcriptomics technology. https://www.seekgene.com/seekspacdxbkjzlzjs, 2025. Accessed: 2025-01-20

    2025

  40. [48]

    Spatial atlas of the mouse central nervous system at molecular resolution

    Shi, H., He, Y., Zhou, Y., Huang, J., Maher, K., Wang, B., Tang, Z., Luo, S., Tan, P., Wu, M., et al. Spatial atlas of the mouse central nervous system at molecular resolution. Nature, 622 0 (7983): 0 552--561, 2023

    2023

  41. [49]

    Shuaibi, M., Kolluru, A., Das, A., Grover, A., Sriram, A., Ulissi, Z., and Zitnick, C. L. Rotation invariant graph neural networks using spin convolutions. arXiv preprint arXiv:2106.09575, 2021

    2021 arXiv

  42. [50]

    L., Salm \'e n, F., Vickovic, S., Lundmark, A., Navarro, J

    St hl, P. L., Salm \'e n, F., Vickovic, S., Lundmark, A., Navarro, J. F., Magnusson, J., Giacomello, S., Asp, M., Westholm, J. O., Huss, M., et al. Visualization and analysis of gene expression in tissue sections by spatial transcriptomics. Science, 353 0 (6294): 0 78--82, 2016

    2016

  43. [51]

    R., Murray, E., Kumar, P., Li, J., Marshall, J

    Stickels, R. R., Murray, E., Kumar, P., Li, J., Marshall, J. L., Di Bella, D. J., Arlotta, P., Macosko, E. Z., and Chen, F. Highly sensitive spatial transcriptomics at near-cellular resolution with slide-seqv2. Nature biotechnology, 39 0 (3): 0 313--319, 2021

    2021

  44. [52]

    Svensson, V., Vento-Tormo, R., and Teichmann, S. A. Exponential scaling of single-cell rna-seq in the past decade. Nature protocols, 13 0 (4): 0 599--604, 2018

    2018

  45. [53]

    Sza ata, A., Hrovatin, K., Becker, S., Tejada-Lapuerta, A., Cui, H., Wang, B., and Theis, F. J. Transformers in single-cell omics: a review and new perspectives. Nature methods, 21 0 (8): 0 1430--1443, 2024

    2024

  46. [54]

    Tejada-Lapuerta, A., Bertin, P., Bauer, S., Aliee, H., Bengio, Y., and Theis, F. J. Causal machine learning for single-cell genomics. Nature Genetics, pp.\ 1--12, 2025

    2025

  47. [55]

    V., Xiao, L., Chopra, A., Chaffin, M

    Theodoris, C. V., Xiao, L., Chopra, A., Chaffin, M. D., Sayed, Z. R. A., Hill, M. C., Mantineo, H., Brydon, E. M., Zeng, Z., Liu, X. S., and Ellinor, P. T. Transfer learning enables predictions in network biology. Nature, 618: 0 616--624, 2023. URL https://api.semanticscholar....

    2023

  48. [56]

    A., Waltman, L., and Van Eck, N

    Traag, V. A., Waltman, L., and Van Eck, N. J. From louvain to leiden: guaranteeing well-connected communities. Scientific reports, 9 0 (1): 0 1--12, 2019

    2019

  49. [57]

    Data release program

    Vizgen. Data release program. https://vizgen.com/data-release-program/, 2025. Accessed: 2025-01-20

    2025

  50. [58]

    M., and Ma, T

    Wei, C., Xie, S. M., and Ma, T. Why do pretrained language models help in downstream tasks? an analysis of head and prompt tuning. Advances in Neural Information Processing Systems, 34: 0 16158--16170, 2021

    2021

  51. [59]

    Cellplm: pre-training of cell language model beyond single cells

    Wen, H., Tang, W., Dai, X., Ding, J., Jin, W., Xie, Y., and Tang, J. Cellplm: pre-training of cell language model beyond single cells. bioRxiv, pp.\ 2023--10, 2023

    2023

  52. [60]

    Perteval-scfm: Benchmarking single-cell foundation models for perturbation effect prediction

    Wenteler, A., Occhetta, M., Branson, N., Huebner, M., Curean, V., Dee, W., Connell, W., Hawkins-Hooker, A., Chung, P., Ektefaie, Y., et al. Perteval-scfm: Benchmarking single-cell foundation models for perturbation effect prediction. bioRxiv, pp.\ 2024--10, 2024

    2024

  53. [61]

    G., Lee, H

    Williams, C. G., Lee, H. J., Asatsuma, T., Vento-Tormo, R., and Haque, A. An introduction to spatial transcriptomics for biomedical research. Genome Medicine, 14 0 (1): 0 68, 2022

    2022

  54. [62]

    A spatiotemporal atlas of cholestatic injury and repair in mice

    Wu, B., Shentu, X., Nan, H., Guo, P., Hao, S., Xu, J., Shangguan, S., Cui, L., Cen, J., Deng, Q., et al. A spatiotemporal atlas of cholestatic injury and repair in mice. Nature Genetics, pp.\ 1--15, 2024

    2024

  55. [63]

    Z., Al-Eryani, G., Roden, D

    Wu, S. Z., Al-Eryani, G., Roden, D. L., Junankar, S., Harvey, K., Andersson, A., Thennavan, A., Wang, C., Torpy, J. R., Bartonicek, N., et al. A single-cell and spatially resolved atlas of human breast cancers. Nature genetics, 53 0 (9): 0 1334--1347, 2021

    2021

  56. [64]

    How powerful are graph neural networks? arXiv preprint arXiv:1810.00826, 2018

    Xu, K., Hu, W., Leskovec, J., and Jegelka, S. How powerful are graph neural networks? arXiv preprint arXiv:1810.00826, 2018

    2018 arXiv

  57. [65]

    Stomicsdb: a comprehensive database for spatial transcriptomics data sharing, analysis and visualization

    Xu, Z., Wang, W., Yang, T., Li, L., Ma, X., Chen, J., Wang, J., Huang, Y., Gould, J., Lu, H., et al. Stomicsdb: a comprehensive database for spatial transcriptomics data sharing, analysis and visualization. Nucleic acids research, 52 0 (D1): 0 D1053--D1061, 2024

    2024

  58. [66]

    Scbert as a large-scale pretrained deep language model for cell type annotation of single-cell rna-seq data

    Yang, F., Wang, W., Wang, F., Fang, Y., Tang, D., Huang, J., Lu, H., and Yao, J. Scbert as a large-scale pretrained deep language model for cell type annotation of single-cell rna-seq data. Nature Machine Intelligence, 4 0 (10): 0 852–866, 2022. doi:10.1038/s42256-022-00534-z

    2022 doi

  59. [67]

    Genecompass: deciphering universal gene regulatory mechanisms with a knowledge-informed cross-species foundation model

    Yang, X., Liu, G., Feng, G., Bu, D., Wang, P., Jiang, J., Chen, S., Yang, Q., Miao, H., Zhang, Y., et al. Genecompass: deciphering universal gene regulatory mechanisms with a knowledge-informed cross-species foundation model. Cell Research, pp.\ 1--16, 2024

    2024

  60. [68]

    Do transformers really perform badly for graph representation? Advances in neural information processing systems, 34: 0 28877--28888, 2021

    Ying, C., Cai, T., Luo, S., Zheng, S., Ke, G., He, D., Shen, Y., and Liu, T.-Y. Do transformers really perform badly for graph representation? Advances in neural information processing systems, 34: 0 28877--28888, 2021

    2021

  61. [69]

    Hierarchical graph representation learning with differentiable pooling

    Ying, Z., You, J., Morris, C., Ren, X., Hamilton, W., and Leskovec, J. Hierarchical graph representation learning with differentiable pooling. Advances in neural information processing systems, 31, 2018

    2018

  62. [70]

    Cell-ontology guided transcriptome foundation model

    Yuan, X., Zhan, Z., Zhang, Z., Zhou, M., Zhao, J., Han, B., Li, Y., and Tang, J. Cell-ontology guided transcriptome foundation model. arXiv preprint arXiv:2408.12373, 2024

    2024 arXiv

  63. [71]

    Q., and Yao, J

    Yuan, Z., Pan, W., Zhao, X., Zhao, F., Xu, Z., Li, X., Zhao, Y., Zhang, M. Q., and Yao, J. Sodb facilitates comprehensive exploration of spatial omics data. Nature Methods, 20 0 (3): 0 387--399, 2023

    2023

  64. [72]

    Large-scale cell representation learning via divide-and-conquer contrastive learning

    Zhao, S., Zhang, J., and Nie, Z. Large-scale cell representation learning via divide-and-conquer contrastive learning. arXiv preprint arXiv:2306.04371, 2023

    2023 arXiv

  65. [73]

    Langcell: Language-cell pre-training for cell identity understanding

    Zhao, S., Zhang, J., Wu, Y., Luo, Y., and Nie, Z. Langcell: Language-cell pre-training for cell identity understanding. arXiv preprint arXiv:2405.06708, 2024

    2024 arXiv

  66. [74]

    Multi-scale protein language model for unified molecular modeling

    Zheng, K., Long, S., Lu, T., Yang, J., Dai, X., Zhang, M., Nie, Z., Ma, W.-Y., and Zhou, H. Multi-scale protein language model for unified molecular modeling. bioRxiv, pp.\ 2024--03, 2024

    2024

  67. [75]

    Causal-iqa: Towards the generalization of image quality assessment based on causal inference

    Zhong, Y., Wu, X., Zhang, L., Yang, C., and Jiang, T. Causal-iqa: Towards the generalization of image quality assessment based on causal inference. In Forty-first International Conference on Machine Learning, 2024

    2024

  68. [76]

    Ctd-inpainting: Towards the coherence of text-driven inpainting with blended diffusion

    Zhong, Y., Zhao, X., Zhao, G., Chen, B., Hao, F., Zhao, R., He, J., Shi, L., and Zhang, L. Ctd-inpainting: Towards the coherence of text-driven inpainting with blended diffusion. Information Fusion, pp.\ 103163, 2025

    2025

  69. [77]

    Uni-mol: A universal 3d molecular representation learning framework

    Zhou, G., Gao, Z., Ding, Q., Zheng, H., Xu, H., Wei, Z., Zhang, L., and Ke, G. Uni-mol: A universal 3d molecular representation learning framework. 2023

    2023

  70. [78]

    Pianno: a probabilistic framework automating semantic annotation for spatial transcriptomics

    Zhou, Y., He, W., Hou, W., and Zhu, Y. Pianno: a probabilistic framework automating semantic annotation for spatial transcriptomics. Nature Communications, 15 0 (1): 0 2848, 2024

    2024

  71. [79]

    Comparative analysis of single-cell rna sequencing methods

    Ziegenhain, C., Vieth, B., Parekh, S., Reinius, B., Guillaumet-Adkins, A., Smets, M., Leonhardt, H., Heyn, H., Hellmann, I., and Enard, W. Comparative analysis of single-cell rna sequencing methods. Molecular cell, 65 0 (4): 0 631--643, 2017

    2017

  72. [80]

    Zormpas, E., Queen, R., Comber, A., and Cockell, S. J. Mapping the transcriptome: Realizing the full potential of spatial data analysis. Cell, 186 0 (26): 0 5677--5689, 2023

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.