Pith. sign in

REVIEW 4 major objections 6 minor 41 references

Sparser2Sparse: Single-shot Sparser-to-Sparse Learning for Spatial Transcriptomics Imputation with Natural Image Co-learning

T0 review · 4 major / 6 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 A single sparse spatial-transcriptomics slide, co-trained with natural images, is claimed to reconstruct dense gene-expression maps and beat TESLA, BayesSpace, and DIST on eight Xenium samples.

desk verdict A genuinely new single-sample training scheme for ST imputation, but the main claim currently hinges on whether the reported metrics include the observed spots that the DC layer trivially preserves. read the letter →

arxiv 2507.16886 v1 pith:XVWQ4MPB submitted 2025-07-22 cs.CV cs.AI

Yaoyu Fang , Jiahe Qian , Xinkun Wang , Lee A. Cooper , Bo Zhou This is my paper · ORCID
classification cs.CVcs.AI
keywords single-shotlearningspatialtranscriptomicsgeneexpressionimputationself-supervisedcross-domainco-trainingdataconsistencyXeniumnaturalimages
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

This paper tries to establish that high-resolution spatial transcriptomics can be reconstructed from a single, cheaply acquired sparse sample rather than from costly dense sequencing. Its method, S2S-ST, trains a per-sample network on one sparsely sampled tissue slide while co-training on ordinary grayscale natural images, using a 'sparser-to-sparse' self-supervised loop plus a cascaded network that repeatedly reinserts the measured spots. The paper reports that on eight Xenium samples across breast, liver, lymphoid, and prostate tissue, this single-shot reconstruction beats TESLA, BayesSpace, and DIST in mean absolute error, Pearson correlation, and structural similarity. A sympathetic reader would care because the high cost and low availability of dense ST data are the main barriers to wider use; if the claim holds, experiments could sequence a fraction of spots and still obtain full-resolution maps.

What carries the argument

The load-bearing mechanism is the sparser-to-sparse self-supervised objective (Eqs. 3-4). From one sparse ST slide, the method subsamples the already-sparse spots to make a 'sparser' input, trains the network to recover the sparse input, and applies the same network to recover the dense map from the sparse input, while the same masks are applied to grayscale natural images for fully supervised co-training (Eqs. 8-9). The Cascaded Data Consistent Imputation Network (CDCIN) then iterates K times: a Data Consistency layer (Eq. 11) reinserts measured spot values into the current prediction before a Residual Dense Hybrid Attention Network refines it, so the known data anchor every stage.

What would settle it

Take any of the eight Xenium datasets, mask out a held-out set of spots not used in the loss, train CDCIN on the remaining sparse spots, and evaluate MAE, PCC, and SSIM on the held-out spots only. Also rerun with synthetic noise or dropout added to the observed spots; if held-out error is much larger than the reported whole-map metrics, or if added noise degrades CDCIN more than it degrades DIST, then hard reinsertion of observed values is carrying the reported advantage.

Watch

Extended reading notes

Core claim

The central claim is that spatial gene-expression maps have enough intrinsic structure, and share enough low-level spatial structure with natural images, that a network never shown a dense ST map can learn to fill missing spots from one sparse slide. The paper's Cascaded Data Consistent Imputation Network (CDCIN) carries this out: each of three stages predicts the full map, a Data Consistency layer swaps predicted values back to measured values wherever spots were sampled, and the refined output is fed to the next stage. On the eight reported Xenium datasets the final predictions achieve lower MAE and higher PCC and SSIM than all three baselines, with ablations attributing consistent gains to both natural-image co-learning and the cascade.

Load-bearing premise

The method assumes the sparse spots it keeps are exact, noise-free measurements and forces them back into the reconstruction at every cascade stage, so if real sparse ST data contains substantial dropout or measurement noise, the reported accuracy could be inflated by the preserved known spots.

Editorial extensions

If this is right

  • If correct, a lab could buy a sparse Xenium run instead of a dense one and obtain a full-resolution expression map for that sample, cutting per-experiment cost.
  • The method's single-shot design removes the need for large collections of dense ST data, which are scarce and often proprietary; only widely available natural images are required for co-training.
  • The reported gains span normal and diseased tissue types, suggesting the learned spatial prior is not tissue-specific.
  • Ablations imply that both natural-image co-training and the data-consistency cascade contribute independently, so future ST imputation networks can build on either component.

Reading between the lines

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

  • A direct test of the natural-image transfer claim would train the same network on ST data only and compare with co-training; larger gains on irregularly textured genes would support the structural-analogy hypothesis.
  • The reported metrics may overstate reconstruction on unseen spots because the Data Consistency layer preserves measured spots in the output; an evaluation restricted to masked-out spots, or with noise injected into observed values, would separate true imputation from anchoring.
  • An implicit corollary is that gene-expression fields with abrupt, sparse, or punctate patterns (for example, rare transcripts) are where this method would be most likely to fail, since natural-image priors favor smoothness and texture continuity.
  • The framework could be extended to cross-sample transfer: train on one tissue's sparse slide and test on another organ without retraining, which the paper leaves for future meta-learning and distillation.
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 / 6 minor

Summary. The paper proposes S2S-ST, a per-sample, single-shot framework for spatial transcriptomics imputation from sparsely sampled high-resolution spots. The method trains a shared cascaded network (CDCIN) jointly on a single ST sample via sparser-to-sparse self-supervised losses and on grayscale natural images via fully supervised losses. The CDCIN architecture repeatedly applies a data-consistency layer that reinserts observed spot values, followed by a Residual Dense Hybrid Attention Network. The authors evaluate on eight Xenium samples from HEST-1k (breast, liver, prostate, lymphoid), one gene per sample, comparing against TESLA, BayesSpace, and DIST with MAE, PCC, and SSIM, and report consistent improvements in Table 1. Ablations study the contribution of natural-image co-training, the data-consistency cascade, cascade depth, and the hybrid attention block.

Significance. If the central claim is validated, S2S-ST addresses a real practical problem: reducing the number of measured spots required for high-resolution ST while retaining reconstruction quality. A notable strength is that the training losses are genuinely self-supervised on the ST side: Eqs. (3) and (4) compare downsampled predictions to the observed sparse inputs, not to dense ground truth, and the natural-image branch provides external supervision. The single-shot, per-sample design is also responsive to the scarcity of public ST data. However, the current evidence is not yet sufficient for the abstract's claim of 'outperforming state-of-the-art in imputation accuracy': the evaluation protocol leaves open whether observed spots are included in the metrics, no variance information is provided, and the cascade depth is tuned on the same samples used for the headline comparison. The paper does partially acknowledge its limitations (per-sample training cost, single-gene scope, lack of downstream biological validation) in the Discussion, which is a positive sign.

major comments (4)
  1. [Sec. 2.6, Eq. (11), Table 1] The evaluation protocol never states whether the MAE/PCC/SSIM values in Table 1 are computed over all spots or only over spots that were unobserved after downsampling. Because the Data Consistency layer in Eq. (11) replaces the prediction at every sampled position with the exact observed value at every cascade, any metric that includes the sampled spots gives CDCIN a trivial identity advantage that TESLA, BayesSpace, and DIST do not share; the reported gains would then partly measure reconstruction of known values rather than imputation of missing spots. Please report metrics separately on the masked-out (unobserved) spots, and state explicitly that the same masked-out evaluation is applied to all baselines.
  2. [Tables 1-2, Figs. 4-5] No error bars, confidence intervals, or repeated-run statistics are reported, and each sample is evaluated on a single gene. The single numbers in Tables 1 and 2 and Figures 4 and 5 are insufficient to support the abstract's claim of consistent superiority, especially because training involves random crops, flips, rotations, and 3000 epochs of stochastic optimization. Please provide multiple seeds with mean and standard deviation, and ideally a paired statistical test, across at least the Table 1 datasets; also clarify how the single gene per tissue was selected and whether the results generalize to other genes.
  3. [Sec. 2.5, Fig. 5, Table 1] The cascade depth K=3 is selected using the same datasets that later appear in the main comparison: Section 2.5 refers to an evaluation in Section 3, and Figure 5A sweeps K on TENX94, TENX96, and TENX97, which are also in Table 1. This creates a selection-on-the-test-set effect and may inflate the reported performance. Please either use a separate validation sample or gene for model selection, or report the full K sweep for all test samples so the reader can judge the sensitivity.
  4. [Sec. 2.2, Eqs. (3)-(4), Eq. (11)] The self-supervised losses and the Data Consistency operation treat every observed sparse value as exact ground truth. Real Xenium measurements contain dropout and technical noise; if the observed spots are noisy, hard reinjection propagates that noise into the reconstruction, and the reported metrics may not reflect performance under realistic measurement error. Please add a sensitivity analysis with synthetic noise or dropout added to the sampled spots, or relax the hard DC constraint to a soft consistency term, and discuss the implications for the claimed cost-reduction benefit.
minor comments (6)
  1. [Sec. 3, Table 1] The text says CDCIN 'can consistently surpass these existing approaches, delivering the lowest MAE and highest SSIM,' but on TENX98 the PCC of CDCIN (0.7992) is lower than that of DIST (0.8085). Please soften the wording or explain this discrepancy.
  2. [Table 1] Each dataset is evaluated on exactly one gene (ERBB2, CYP2A7, XBP1, CCN1). A statement of how these genes were selected, together with multi-gene results or a justification for why one gene per sample is representative, would considerably strengthen the imputation claim.
  3. [Table 1] No simple interpolation or smoothing baseline (e.g., bilinear interpolation or k-nearest-neighbor imputation) is included. Such a baseline would help calibrate the magnitude of the reported improvements over TESLA, BayesSpace, and DIST.
  4. [General] The manuscript does not state whether code and trained models will be released. A data/code availability statement would significantly aid reproducibility.
  5. [General] There are several typos and ligature artifacts (e.g., 'total loos' in Section 2.1, 'o ffering' in the Introduction, 'e ffective' and 'di fficult' elsewhere). A careful proofread is needed.
  6. [Fig. 5] The MAE curves in Figure 5 are described without error bars or clear axis labels in the text; adding standard deviations and labeled axes would improve interpretability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: training losses are self-supervised consistency objectives, the DC layer is an explicit data-preservation mechanism rather than a disguised fit, and all cited inputs are external to the paper's own claims.

full rationale

The S2S-ST derivation chain is self-contained. The ST self-supervised losses (Eqs. 3-4) compare the network's predictions, after downsampling, to the same sparse observations used as input; this is a consistency objective, not a fitted parameter renamed as a prediction, and dense ground truth is never used as a training target in the ST branch. The natural-image branch (Eqs. 8-9) provides genuinely external dense supervision from DIV2K. The DC layer (Eq. 11) re-injects observed spot values at mask positions by construction, so those positions are not 'predicted' in the sense of being inferred from other data, but the paper explicitly frames this as preserving sampled data fidelity, and the missing-spot imputation is not forced by this operation. There are no load-bearing self-citations: the cited baselines, datasets (HEST-1K, DIV2K), and architectural components (Swin, HAT/CAB scaling) are all external to the present authors' prior work. A separate evaluation concern--whether Table 1 metrics include or exclude the re-injected observed spots--is a soundness or benchmarking question, not circular reasoning, because no equation reduces the reported imputation accuracy to the training inputs without independent evaluation. Therefore no circular step is exhibited.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests mainly on domain assumptions about transferability between natural images and gene expression, noiselessness of observed spots, and the evaluation protocol. No new physical entities are introduced. The main free hyperparameters are lambda, alpha, and the cascade depth chosen on the test datasets.

free parameters (3)
  • gene loss weight lambda = 10
    Section 2.5: balances ST and GNI losses; set by hand, not swept across values.
  • channel attention scale alpha = 0.01
    Section 2.5: scales CAB output; taken from prior HAT work but remains a chosen hyperparameter.
  • cascade depth K = 3
    Section 2.5 and Fig. 5A: selected by ablation on TENX94/96/97, which are the same datasets used for final reported results.
assumptions (4)
  • domain assumption Gene expression spatial maps and grayscale natural images share sufficiently similar local structure that co-training on natural images improves ST imputation.
    Section 2.1 and Discussion claim structural analogies between gene expression patterns and natural textures; this is the basis for the GNI co-training loss (Eqs. 6-10).
  • domain assumption Sparse ST spots are observed exactly without noise; the Data Consistency layer and self-supervised losses treat them as ground truth anchors.
    Eq. (11) reintroduces observed values directly; Eqs. (3)-(4) penalize deviations at sampled positions. Real ST measurements contain dropout and noise, which would be propagated.
  • domain assumption The deterministic regular-grid downsampling mask (top-left pixel of each 2x2 block) is representative of practical sparse sampling in ST.
    Section 2.5: downsampling by selecting the top-left pixel of each SxS grid; no irregular or random mask experiments are reported.
  • domain assumption Evaluation metrics computed on the reconstructed map reflect imputation quality; the paper does not clarify whether the preserved sampled spots are included in MAE/PCC/SSIM.
    If included, the DC layer exactly reproduces known spots, artificially lowering MAE and raising SSIM for CDCIN relative to baselines that do not hard-preserve inputs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sparser2Sparse: Single-shot Sparser-to-Sparse Learning for Spatial Transcriptomics Imputation with Natural Image Co-learning." pith.science (2026). https://pith.science/paper/XVWQ4MPB

@misc{pith2026250716886,
  author       = {Pith},
  title        = {Pith review of: Sparser2Sparse: Single-shot Sparser-to-Sparse Learning for Spatial Transcriptomics Imputation with Natural Image Co-learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XVWQ4MPB}},
  note         = {Machine review of arXiv:2507.16886}
}
read the original abstract

Spatial transcriptomics (ST) has revolutionized biomedical research by enabling high resolution gene expression profiling within tissues. However, the high cost and scarcity of high resolution ST data remain significant challenges. We present Single-shot Sparser-to-Sparse (S2S-ST), a novel framework for accurate ST imputation that requires only a single and low-cost sparsely sampled ST dataset alongside widely available natural images for co-training. Our approach integrates three key innovations: (1) a sparser-to-sparse self-supervised learning strategy that leverages intrinsic spatial patterns in ST data, (2) cross-domain co-learning with natural images to enhance feature representation, and (3) a Cascaded Data Consistent Imputation Network (CDCIN) that iteratively refines predictions while preserving sampled gene data fidelity. Extensive experiments on diverse tissue types, including breast cancer, liver, and lymphoid tissue, demonstrate that our method outperforms state-of-the-art approaches in imputation accuracy. By enabling robust ST reconstruction from sparse inputs, our framework significantly reduces reliance on costly high resolution data, facilitating potential broader adoption in biomedical research and clinical applications.

Figures

Figures reproduced from arXiv: 2507.16886 by the authors.

Figure 1
Figure 1. Overview of the Single-shot S2S-ST framework. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Detailed architecture of the CDCIN used in our S2S-ST framework in Fig. 1. The model adopts a cascaded structure composed of [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Qualitative comparison of spatial gene expression reconstruction for the ERBB2 gene across three breast cancer samples (TENX94, [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Ablation study evaluating the impact of key components in the proposed CDCIN model using ERBB2 expression in breast cancer [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Ablation study on the effect of cascade depth and the Hybrid Attention Block (HAB) in the CDCIN model. (A) MAE comparison across different numbers of cascaded stages (from 1 to 5) on datasets TENX94, TENX96, and TENX97. Performance improves significantly up to three ca…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 33 canonical work pages

  1. [1]

    , author Timofte, R

    author Agustsson, E. , author Timofte, R. , year 2017 . title NTIRE 2017 Challenge on Single Image Super-Resolution : Dataset and Study , in: booktitle 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops ( CVPRW ) , pp. pages 1122--1131 . :10.1109/CVPRW.2017.150

    doi:10.1109/cvprw.2017.150 2017

  2. [2]

    , author Song, A.H

    author Almagro-P \'e rez , C. , author Song, A.H. , author Weishaupt, L. , author Kim, A. , author Jaume, G. , author Williamson, D.F.K. , author Hemker, K. , author Lu, M.Y. , author Singh, K. , author Chen, B. , author Le, L.P. , author Baras, A.S. , author Jiang, S. , author Bashashati, A. , author Liu, J.T.C. , author Mahmood, F. , year 2025 . title A...

    arXiv doi:10.48550/arxiv.2502.17761 2025

  3. [3]

    , author Lu, Y

    author An, J. , author Lu, Y. , author Chen, Y. , author Chen, Y. , author Zhou, Z. , author Chen, J. , author Peng, C. , author Huang, R. , author Peng, F. , year 2024 . title Spatial transcriptomics in breast cancer: Providing insight into tumor heterogeneity and promoting individualized therapy . journal Frontiers in Immunology volume 15 . :10.3389/fim...

    work page arXiv 2024

  4. [4]

    , author Larsson, L

    author Andersson, A. , author Larsson, L. , author Stenbeck, L. , author Salm \'e n, F. , author Ehinger, A. , author Wu, S.Z. , author Al-Eryani , G. , author Roden, D. , author Swarbrick, A. , author Borg, . , author Fris \'e n, J. , author Engblom, C. , author Lundeberg, J. , year 2021 . title Spatial deconvolution of HER2-positive breast cancer deline...

    doi:10.1038/s41467-021-26271-2 2021

  5. [5]

    , author Bergenstr hle, J

    author Asp, M. , author Bergenstr hle, J. , author Lundeberg, J. , year 2020 . title Spatially Resolved Transcriptomes --- Next Generation Tools for Tissue Exploration . journal BioEssays volume 42 , pages 1900221 . :10.1002/bies.201900221

    work page doi:10.1002/bies.201900221 2020

  6. [6]

    , author Warrell, J

    author Aung, T.N. , author Warrell, J. , author Martinez-Morilla , S. , author Gavrielatou, N. , author Vathiotis, I. , author Yaghoobi, V. , author Kluger, H.M. , author Gerstein, M. , author Rimm, D.L. , year 2024 . title Spatially Informed Gene Signatures for Response to Immunotherapy in Melanoma . journal Clinical Cancer Research volume 30 , pages 352...

    work page doi:10.1158/1078-0432.ccr-23-3932 2024

  7. [7]

    , author He, B

    author Bergenstr hle, L. , author He, B. , author Bergenstr hle, J. , author Abalo, X. , author Mirzazadeh, R. , author Thrane, K. , author Ji, A.L. , author Andersson, A. , author Larsson, L. , author Stakenborg, N. , author Boeckxstaens, G. , author Khavari, P. , author Zou, J. , author Lundeberg, J. , author Maaskola, J. , year 2022 . title Super-resol...

    work page doi:10.1038/s41587-021-01075-3 2022

  8. [8]

    , author Triandafillou, C.G

    author Boe, R.H. , author Triandafillou, C.G. , author Lazcano, R. , author Wargo, J.A. , author Raj, A. , year 2024 . title Spatial transcriptomics reveals influence of microenvironment on intrinsic fates in melanoma therapy resistance . :10.1101/2024.06.30.601416

    work page doi:10.1101/2024.06.30.601416 2024

Show all 41 references
  1. [9]

    , author Li, X

    author Boileau, E. , author Li, X. , author Naarmann-de Vries , I.S. , author Becker, C. , author Casper, R. , author Altm \"u ller, J. , author Leuschner, F. , author Dieterich, C. , year 2022 . title Full- Length Spatial Transcriptomics Reveals the Unexplored Isoform Diversi...

    2022

  2. [10]

    , author Chen, C

    author Chen, R.J. , author Chen, C. , author Li, Y. , author Chen, T.Y. , author Trister, A.D. , author Krishnan, R.G. , author Mahmood, F. , year 2022 . title Scaling Vision Transformers to Gigapixel Images via Hierarchical Self-Supervised Learning . :10.48550/arXiv.2206.0264...

    2022 doi arXiv

  3. [11]

    , author Lu, A

    author Chen, W.T. , author Lu, A. , author Craessaerts, K. , author Pavie, B. , author Sala Frigerio, C. , author Corthout, N. , author Qian, X. , author Lal \'a kov \'a , J. , author K \"u hnemund, M. , author Voytyuk, I. , author Wolfs, L. , author Mancuso, R. , author Salta...

    2020

  4. [12]

    , author Wang, X

    author Chen, X. , author Wang, X. , author Zhou, J. , author Qiao, Y. , author Dong, C. , year 2023 . title Activating More Pixels in Image Super-Resolution Transformer , in: booktitle 2023 IEEE / CVF Conference on Computer Vision and Pattern Recognition ( CVPR ) , publisher I...

    2023

  5. [13]

    , author Cockenpot, V

    author Coutant, A. , author Cockenpot, V. , author Muller, L. , author Degletagne, C. , author Pommier, R. , author Tonon, L. , author Ardin, M. , author Michallet, M.C. , author Caux, C. , author Laurent, M. , author Morel, A.P. , author Saintigny, P. , author Puisieux, A. , ...

    2023

  6. [14]

    , author Chen, J

    author Dries, R. , author Chen, J. , author del Rossi, N. , author Khan, M.M. , author Sistig, A. , author Yuan, G.C. , year 2021 . title Advances in spatial transcriptomic data analysis . journal Genome Research volume 31 , pages 1706--1718 . :10.1101/gr.275224.121

    2021 doi

  7. [15]

    , author Yang, Y.C

    author Du, J. , author Yang, Y.C. , author An, Z.J. , author Zhang, M.H. , author Fu, X.H. , author Huang, Z.F. , author Yuan, Y. , author Hou, J. , year 2023 . title Advances in spatial transcriptomics and related data analysis strategies . journal Journal of Translational Me...

    2023 doi

  8. [16]

    , author Chen, B

    author Fang, S. , author Chen, B. , author Zhang, Y. , author Sun, H. , author Liu, L. , author Liu, S. , author Li, Y. , author Xu, X. , year 2023 . title Computational Approaches and Challenges in Spatial Transcriptomics . journal Genomics, Proteomics & Bioinformatics volume...

    2023 doi

  9. [17]

    , year 2021

    author Gevaert, O. , year 2021 . title Meta-learning reduces the amount of data needed to build AI models in oncology . journal British Journal of Cancer volume 125 , pages 309--310 . :10.1038/s41416-021-01358-1

    2021 doi

  10. [18]

    , author Lu, W

    author He, Y. , author Lu, W. , author Zhou, X. , author Mu, J. , author Shen, W. , year 2024 . title Unraveling Alzheimer 's disease: Insights from single-cell sequencing and spatial transcriptomic . journal Frontiers in Neurology volume 15 . :10.3389/fneur.2024.1515981

    2024

  11. [19]

    , author Coleman, K

    author Hu, J. , author Coleman, K. , author Zhang, D. , author Lee, E.B. , author Kadara, H. , author Wang, L. , author Li, M. , year 2023 . title Deciphering tumor ecosystems at super resolution from spatial transcriptomics with TESLA . journal Cell Systems volume 14 , pages ...

    2023 doi

  12. [20]

    , author Doucet, P

    author Jaume, G. , author Doucet, P. , author Song, A.H. , author Lu, M.Y. , author Almagro-P \'e rez , C. , author Wagner, S.J. , author Vaidya, A.J. , author Chen, R.J. , author Williamson, D.F.K. , author Kim, A. , author Mahmood, F. , year 2024 . title HEST-1k : A Dataset ...

    2024 doi arXiv

  13. [21]

    , author Kim, T.K

    author Jung, N. , author Kim, T.K. , year 2023 . title Spatial transcriptomics in neuroscience . journal Experimental & Molecular Medicine volume 55 , pages 2105--2115 . :10.1038/s12276-023-01093-y

    2023 doi

  14. [22]

    , author Frolovait \.e , P

    author Kleino, I. , author Frolovait \.e , P. , author Suomi, T. , author Elo, L.L. , year 2022 . title Computational solutions for spatial transcriptomics . journal Computational and Structural Biotechnology Journal volume 20 , pages 4870--4884 . :10.1016/j.csbj.2022.08.043

    2022 doi

  15. [23]

    u nnemann, F. , author Nagai, J.S. , author Halder, M. , author Schumacher, D. , author Menzel, S. , author Sch \

    author Kuppe, C. , author Ramirez Flores, R.O. , author Li, Z. , author Hayat, S. , author Levinson, R.T. , author Liao, X. , author Hannani, M.T. , author Tanevski, J. , author W \"u nnemann, F. , author Nagai, J.S. , author Halder, M. , author Schumacher, D. , author Menzel,...

    2022

  16. [24]

    , author Tekpli, X

    author Levy-Jurgenson , A. , author Tekpli, X. , author Kristensen, V.N. , author Yakhini, Z. , year 2020 . title Spatial transcriptomics inferred from pathology whole-slide images links tumor heterogeneity to survival in breast and lung cancer . journal Scientific Reports vol...

    2020 doi

  17. [25]

    , author Zhang, X

    author Li, Q. , author Zhang, X. , author Ke, R. , year 2022 . title Spatial Transcriptomics for Tumor Heterogeneity Analysis . journal Frontiers in Genetics volume 13 . :10.3389/fgene.2022.906158

    2022

  18. [26]

    , author Yuan, X

    author Long, X. , author Yuan, X. , author Du, J. , year 2023 . title Single-cell and spatial transcriptomics: Advances in heart development and disease applications . journal Computational and Structural Biotechnology Journal volume 21 , pages 2717--2731 . :10.1016/j.csbj.2023.04.007

    2023 doi

  19. [27]

    , author Koido, M

    author Monjo, T. , author Koido, M. , author Nagasawa, S. , author Suzuki, Y. , author Kamatani, Y. , year 2022 . title Efficient prediction of a spatial transcriptomics profile better characterizes breast cancer tissue sections without costly experimentation . journal Scienti...

    2022 doi

  20. [28]

    , author Briskina, A

    author Moslemi, A. , author Briskina, A. , author Dang, Z. , author Li, J. , year 2024 . title A survey on knowledge distillation: Recent advancements . journal Machine Learning with Applications volume 18 , pages 100605 . :10.1016/j.mlwa.2024.100605

    2024

  21. [29]

    , author Croteau, D.L

    author Navarro, J.F. , author Croteau, D.L. , author Jurek, A. , author Andrusivova, Z. , author Yang, B. , author Wang, Y. , author Ogedegbe, B. , author Riaz, T. , author St en, M. , author Desler, C. , author Rasmussen, L.J. , author T njum, T. , author Galas, M.C. , author...

    2020

  22. [30]

    , author Rajewsky, N

    author Piwecka, M. , author Rajewsky, N. , author Rybak-Wolf , A. , year 2023 . title Single-cell and spatial transcriptomics: Deciphering brain complexity in health and disease . journal Nature Reviews. Neurology volume 19 , pages 346--362 . :10.1038/s41582-023-00809-y

    2023 doi

  23. [31]

    , author Kim, S

    author Roth, R. , author Kim, S. , author Kim, J. , author Rhee, S. , year 2020 . title Single-cell and spatial transcriptomics approaches of cardiovascular development and disease . journal BMB reports volume 53 , pages 393--399 . :10.5483/BMBRep.2020.53.8.130

    2020 doi

  24. [32]

    , author Zhou, H.M

    author Sankar, M. , author Zhou, H.M. , author Wang, Q. , author Liu, Y. , author Xuei, X. , author Gao, H. , author Alomari, A. , author Warren, S. , author Turner, M. , year 2024 . title 50480 Immune Infiltrate Profiling of Cutaneous Melanoma with Spatial Transcriptomics . j...

    2024 doi

  25. [33]

    , author Prince, D.K

    author Smith, K.D. , author Prince, D.K. , author MacDonald, J.W. , author Bammler, T.K. , author Akilesh, S. , year 2024 . title Challenges and Opportunities for the Clinical Translation of Spatial Transcriptomics Technologies . journal Glomerular Diseases volume 4 , pages 49...

    2024 doi

  26. [34]

    , author Broadbent, C

    author Song, T. , author Broadbent, C. , author Kuang, R. , year 2023 . title GNTD : Reconstructing spatial transcriptomes with graph-guided neural tensor decomposition informed by spatial and functional relations . journal Nature Communications volume 14 , pages 8276 . :10.10...

    2023 doi

  27. [35]

    , author Li, X

    author Wang, N. , author Li, X. , author Wang, R. , author Ding, Z. , year 2021 . title Spatial transcriptomics and proteomics technologies for deconvoluting the tumor microenvironment . journal Biotechnology Journal volume 16 , pages 2100041 . :10.1002/biot.202100041

    2021 doi

  28. [36]

    , author Hossain, M.Z

    author Yang, Y. , author Hossain, M.Z. , author Stone, E.A. , author Rahman, S. , year 2022 . title Exemplar Guided Deep Neural Network for Spatial Transcriptomics Analysis of Gene Expression Prediction . :10.48550/arXiv.2210.16721, arXiv:2210.16721 http://arxiv.org/abs/2210.16721

    2022 doi arXiv

  29. [37]

    , author Schroeder, A

    author Zhang, D. , author Schroeder, A. , author Yan, H. , author Yang, H. , author Hu, J. , author Lee, M.Y.Y. , author Cho, K.S. , author Susztak, K. , author Xu, G.X. , author Feldman, M.D. , author Lee, E.B. , author Furth, E.E. , author Wang, L. , author Li, M. , year 202...

    2024 doi

  30. [38]

    , author Geng, Y

    author Zhang, S. , author Geng, Y. , author Jiang, X. , author Sun, Z. , author Yan, M. , author Bi, J. , author Tian, X. , author Wang, Q. , year 2024 b. title Investigating the mechanisms of inflammation and immune alterations in Parkinson 's disease using spatial transcript...

    2024

  31. [39]

    , author Stone, M.R

    author Zhao, E. , author Stone, M.R. , author Ren, X. , author Guenthoer, J. , author Smythe, K.S. , author Pulliam, T. , author Williams, S.R. , author Uytingco, C.R. , author Taylor, S.E.B. , author Nghiem, P. , author Bielas, J.H. , author Gottardo, R. , year 2021 . title S...

    2021 doi

  32. [40]

    , author Wang, K

    author Zhao, Y. , author Wang, K. , author Hu, G. , year 2023 . title DIST : Spatial transcriptomics enhancement using deep learning . journal Briefings in Bioinformatics volume 24 , pages bbad013 . :10.1093/bib/bbad013

    2023 doi

  33. [41]

    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 gl...

Pith tools

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