Pith. sign in

REVIEW 4 major objections 6 minor 35 references

VOICE: A Vision-Omics Foundation Model Integrating Direct and Retrieval-Based Prediction of In-situ Single-Cell Gene Expression

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read VOICE predicts single-cell gene expression from H&E images alone, transferring to genes never measured in the target tissue.

desk verdict Solid single-cell expression prediction pipeline with honest ablations; the 'held-out patients' claim needs a patient-level split before the cross-slide numbers can be trusted. read the letter →

arxiv 2608.08366 v1 pith:CWPPJXI2 submitted 2026-08-08 cs.CV q-bio.GN

Xin Luo , Yicheng Tao , Haoxuan Zeng , Suyuan Wang , Chenzi Ouyang , Meiqi Zhu , Kai Liu , Shuibing Chen
show 1 more author
Jie Liu
This is my paper · ORCID
classification cs.CVq-bio.GN
keywords spatialtranscriptomicsH&Ehistologygeneexpressionpredictionfoundationmodelcontrastivealignmentretrieval-augmentedper-genefusionsingle-cell
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

VOICE is a foundation model that predicts a single cell's gene expression purely from its appearance in an H&E-stained image, trained on paired Xenium measurements. The paper's central claim is that this prediction generalizes to held-out patients, slides, and partially overlapping gene panels, and that fusing a direct morphology-to-expression regression with a retrieval branch that copies measured expression from similar reference cells beats either strategy alone. If the claim holds, archived H&E slides—costing tens of dollars and collected routinely—could be mined for single-cell molecular information that currently requires spatial transcriptomics assays costing thousands of dollars per slide. The model also reports predicting genes that were never measured in the target tissue, by transferring morphological knowledge from other tissues.

What carries the argument

The central object is the shared cell-mask-pooled morphology embedding that feeds both prediction branches, together with the per-gene gate that weights their outputs. This embedding comes from a LoRA-adapted pathology encoder aligned to a transcriptome foundation model's space in Stage 1 and refined by an SE(2)-equivariant spatial decoder in Stage 2, so both branches read the same representation while making partly independent errors; the gate picks the more reliable branch gene by gene.

What would settle it

Hold out an entire tissue that appears in no training slide, build the retrieval bank exclusively from other tissues, and measure the direct branch's per-gene Pearson correlation on that tissue's H&E test slide; if the all-gene PCC falls to near zero rather than remaining in the 0.2 range reported for same-tissue genes, the claimed cross-tissue transfer of morphology-to-expression knowledge fails beyond the trained tissue set.

Watch

Extended reading notes

Core claim

VOICE aligns cell-mask-pooled H&E morphology features from a pathology foundation model with single-cell transcriptomic embeddings from a transcriptome foundation model, using LoRA adapters and a symmetric InfoNCE loss over 23 million cells from 15 tissues. From the resulting shared space, a direct branch predicts expression through an SE(2)-equivariant spatial decoder with a shared count head over a 6,029-gene vocabulary, while a retrieval branch averages measured expression from the nearest reference cells in morphology space. A per-gene convex gate then fuses the two predictions, favoring direct regression for genes whose expression tracks visible morphology and favoring retrieval for genes that do not. On seven per-gene Pearson correlation metrics, the model reports the highest scores in all 21 in-slide comparisons and on nearly every cross-slide panel across breast, lung, kidney, ovary, and pancreas, and it predicts genes measured only in other tissues at levels comparable to same-tissue genes.

Load-bearing premise

Retrieval assumes that cells whose H&E appearance is similar also have similar expression for genes with no direct morphological signal, and that a same-tissue reference bank of measured cells exists at inference time; without such a bank, the retrieval branch and its cross-gene recovery stop working.

Editorial extensions

If this is right

  • Archived H&E slides can be re-analyzed as single-cell expression maps without running new spatial assays.
  • Genes absent from a slide's measured panel remain predictable, since the shared count vocabulary is trained across heterogeneous panels.
  • Per-gene gating lets the method be applied to any gene that appears in at least one reference tissue, not just genes with visible morphological correlates.
  • Scaling the training set from 7 million to 23 million cells improved every in-slide metric, suggesting further gains from larger cell corpora.
  • Zero-shot transfer to CosMx slides, though weaker than within-platform performance, indicates the morphology–expression link is not specific to Xenium.

Reading between the lines

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

  • A tissue type entirely absent from the 15 training tissues would be the strictest test: if the direct branch's all-gene PCC collapses, the claimed cross-tissue gene transfer is partly a memorization of the trained tissue distribution.
  • The same-tissue requirement for retrieval banks implies that practical deployment depends on reference-slide availability; building batch-corrected cross-tissue banks and testing whether the per-gene gate transfers across tissues are natural next steps.
  • One could test the per-gene gate's semantics directly: genes with weights near 1 should be enriched for morphological markers like structural proteins, while genes near 0 should be enriched for non-morphological programs such as immune response, which would confirm the mechanism the gate is meant to implement.
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 introduces VOICE, a foundation model that predicts single-cell gene expression from H&E images using paired Xenium spatial transcriptomics data. VOICE first contrastively aligns cell-mask-pooled H&E morphology features from UNI2-h with scFoundation expression embeddings over 23 million cells, then continues fine-tuning with an SE(2)-equivariant spatial decoder and a shared count head over a 6,029-gene vocabulary. At inference, a direct branch regresses expression from morphology, a retrieval branch averages measured expression from the K nearest reference cells under cosine similarity, and a per-gene gate fuses the two predictions with weights fit per tissue. The paper claims that VOICE generalizes to held-out patients, slides, and partially overlapping gene panels, consistently outperforming GHIST, sCellST, DeepSpot2Cell, and DeepSpot-M on seven PCC-based metrics in in-slide and cross-slide benchmarks, and that its direct branch can transfer genes measured only in other tissues. A zero-shot Xenium-to-CosMx transfer is also reported.

Significance. VOICE is a technically strong contribution: it is trained on a large corpus (23M cells, 75 slides, 15 tissues), includes careful ablations of the contrastive objective (Tables 8–9), cell-feature pooling (Table 10), training stages (Table 7), and fusion (Table 6), and it reports an honest cross-platform evaluation with a novel registration procedure. The per-gene fusion of direct and retrieval prediction is a sensible and well-motivated mechanism. If the generalization claims are confirmed, VOICE could enable cost-effective molecular analysis of large H&E archives. However, the lack of a patient-level evaluation and the absence of uncertainty estimates are significant gaps that must be addressed before the central claims can be accepted.

major comments (4)
  1. [Section 4.1, Table 4] The abstract claims that VOICE generalizes to held-out patients, but the evaluation protocol in Section 4.1 only defines slide-level separation: five vertical bands for the in-slide evaluation and 'unseen slides' for the cross-slide evaluation. No patient identifiers are reported, and the paper does not state whether training and test slides come from disjoint patients. Since the retrieval branch constructs its reference bank from training slides of the same tissue (Section 3.3, Eq. 3; Section 4.1), any overlap in patients between the training set and the cross-slide test slides could inflate the reported cross-slide gains for both the retrieval and direct branches. This is a concrete, checkable omission that directly bears on the central generalization claim. Please provide patient-level metadata and verify that the cross-slide test slides are patient-disjoint from all training slides, or re-run the evaluation under an explicit patient-level split.
  2. [Tables 1 and 2] All reported results are point estimates of per-gene Pearson correlation averaged over gene sets, with no confidence intervals, standard errors, or significance tests. For the small gene sets (H20, S20), the mean PCC may have substantial uncertainty, and many of the observed differences between VOICE-23M and the strongest baseline are small (e.g., Table 2, kidney H20 in the Xenium-shared panel: 0.525 vs 0.527). Without measures of uncertainty, the claim of 'consistently outperforms' on all seven metrics is not fully supported. Please report per-gene PCC distributions, bootstrap confidence intervals, or paired tests across genes for the key comparisons.
  3. [Section 3.5, Eq. (7)] The per-gene fusion weights beta_g are selected by grid search on training slides to maximize the correlation between the fused prediction and ground-truth expression, which is the same metric used for the main evaluation. With a small number of training slides per tissue (e.g., pancreas uses four slides in Appendix A.3), this selection procedure may overfit and lead to optimistic fusion gains. The paper should report the variability of beta_g across training slides or select weights using a metric-neutral criterion (e.g., log-likelihood) to demonstrate that the fusion benefit in Table 6 is not an artifact of fitting the evaluation metric.
  4. [Abstract and Section 4.3] The abstract states that VOICE 'consistently outperforms prior single-cell expression prediction methods on seven metrics,' but Table 2 contains a tie (kidney H20 in the full target panel, 0.554 for both VOICE-23M and fine-tuned DeepSpot-M) and a loss (kidney H20 in the Xenium-shared panel, 0.525 vs 0.527). While Section 4.3 acknowledges these exceptions, the abstract overstates the result. Please soften the claim or incorporate the exceptions into the abstract.
minor comments (6)
  1. [Section 3.4] The SE(2)-equivariant transformer is attributed to reference [31], Uni-Mol, which is a 3D molecular representation model, not an SE(2)-equivariant spatial transformer for cell neighborhoods; please provide the correct citation or describe the architecture in the text.
  2. [Section 3.3] Section 3.3 states that K=200 neighbors and tau=0.03 are used in all experiments, but the contrastive-objective experiments in Appendix B.3 use K=20; please clarify which retrieval settings apply to the objective comparison.
  3. [Table 2] Table 2, full target panel, kidney column for DeepSpot-M (fine-tuned) contains run-together values ('0.4790.5540.473'); please fix the table formatting.
  4. [Appendix D.2] Appendix D.2 contains a typo: 'fix the registration. between the H&E and DAPI images' should read 'fix the registration between the H&E and DAPI images.'
  5. [Appendix A.1] The selection criteria for the 24-slide subset used to train VOICE-7M are not described; please state whether the subset is stratified by tissue and how it was chosen.
  6. [Section 1] Section 1: 'which cannot learn morphology–molecular relationships that transfer across datasets' would be clearer as 'which cannot learn transferable morphology–molecular relationships across datasets.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; all reported predictions are evaluated on held-out cells against external ground truth, and no load-bearing step reduces to its own inputs.

full rationale

VOICE is a supervised prediction system, not a derivation, and its central evaluation is against measured expression on cells that are explicitly excluded from the training or reference set. The retrieval branch (Eq. 3) averages measured expression of reference cells, but the paper states the bank excludes the query cell and the query cell's slide, so the predicted quantity (the query cell's expression) never appears in the reference bank by construction. The per-gene fusion weight beta_g (Eq. 7) is fit on training bands or training slides to maximize correlation with measured expression, then fixed and applied to test bands or unseen slides; this is a standard supervised hyperparameter fit, not a test-set oracle. The Stage-1 contrastive objective and Stage-2 count loss use measured expression only as training targets, and the appendix includes ablations against frozen encoders and alternative objectives, so the reported gains are not forced by the architecture choice. No uniqueness theorem, ansatz, or prior claim is imported from the authors' own earlier work; the external foundation models cited (UNI2-h, scFoundation, and the baseline methods) are independent on-paper inputs with their own published support. The abstract's phrase 'held-out patients' is stronger than the slide-level protocol described in Section 4.1, and a same-patient slide in the retrieval bank could inflate cross-slide scores, but that is a data-split validity concern requiring external patient identifiers to check, not a definitional or self-citational circularity. The 'other-tissue-only' gene experiment (Table 3) deliberately uses only the direct branch and tests genes absent from all same-tissue training slides, which is a genuinely held-out generalization claim. I therefore find no circular step that reduces a prediction to its own input; the appropriate score is 0.

Assumptions & free parameters 7 free parameters · 7 assumptions · 0 invented entities

The central claim rests on several domain assumptions about the learnability of expression from morphology, the transferability of fusion weights, and the availability of same-tissue reference banks. The main fitted parameters are the per-gene fusion weights and a set of hand-chosen hyperparameters; no new scientific entities are introduced.

free parameters (7)
  • Per-gene fusion weight beta_g = per-gene, grid search in [0,1], values not reported
    Selected on training slides to maximize correlation between fused prediction and ground truth (Section 3.5, Eq. 7).
  • Retrieval neighbor count K = 200
    Hand-chosen for all retrieval experiments (Section 3.3, Eq. 3).
  • Retrieval temperature tau = 0.03
    Hand-chosen for the similarity weighting in Eq. (3).
  • Contrastive temperature tau_c = 0.07
    Fixed temperature in the Stage 1 InfoNCE loss (Eq. 2).
  • H&E crop size = 55 um field of view, resized to 224x224
    Hand-chosen crop geometry for cell representation (Section 3.2).
  • LoRA configuration = rank 16, alpha 32, dropout 0.05
    Hand-chosen adaptation settings for UNI2-h (Section 3.3).
  • Stage 2 loss coefficients = 1 for log-space MSE, 1/2 for negative-binomial term
    Hand-chosen weighting of the two terms in Eq. (6).
assumptions (7)
  • domain assumption H&E morphology contains transferable signal about single-cell gene expression
    Central premise of the method, stated in Section 1 and used throughout.
  • domain assumption scFoundation embeddings provide a unified transcriptomic latent space across heterogeneous gene panels
    Stage 1 aligns morphology to scFoundation embeddings (Section 3.3).
  • domain assumption UNI2-h features retain morphology information after LoRA adaptation
    Both stages rely on the adapted UNI2-h encoder as the image representation.
  • domain assumption Reference cells with similar morphology have similar expression for retrieval-predicted genes
    The retrieval branch in Eq. (3) averages expression from morphology-similar cells.
  • domain assumption Per-gene fusion weights estimated on training slides transfer to held-out slides of the same tissue
    Stage 3 fits beta_g on training slides and applies it to test slides (Section 3.5).
  • domain assumption The curated Xenium slides and public datasets are representative and free of gross registration errors
    All training and evaluation depend on registered H&E and Xenium pairs (Appendix A.1, A.2).
  • domain assumption Negative-binomial and log-MSE objectives are appropriate for count prediction
    The Stage 2 loss in Eq. (6) combines these two objectives.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VOICE: A Vision-Omics Foundation Model Integrating Direct and Retrieval-Based Prediction of In-situ Single-Cell Gene Expression." pith.science (2026). https://pith.science/paper/CWPPJXI2

@misc{pith2026260808366,
  author       = {Pith},
  title        = {Pith review of: VOICE: A Vision-Omics Foundation Model Integrating Direct and Retrieval-Based Prediction of In-situ Single-Cell Gene Expression},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CWPPJXI2}},
  note         = {Machine review of arXiv:2608.08366}
}
read the original abstract

Spatial transcriptomics can resolve gene expression at single-cell resolution, but it is costly, limited to targeted panels of a few hundred to a few thousand genes, and applicable to only a small number of samples. H&E imaging, by contrast, is cheap and collected routinely at scale. This makes predicting single-cell expression directly from morphology a practical way to bring molecular analysis to large tissue archives. We therefore present VOICE, a multimodal foundation model that predicts single-cell gene expression from H&E images using paired Xenium data. VOICE first aligns cell centered H&E morphology from a pathology foundation model with single-cell expression embeddings from a transcriptome foundation model, trained using contrastive learning over 23 million cells. Next it predicts expression through two branches. One branch directly regresses expression from morphology. The other branch retrieves measured expression from similar reference cells, recovering genes that do not have morphological signal. Because genes vary in morphological predictability, VOICE fuses the two branches with a per-gene weight. After training, VOICE generalizes to heldout patients, slides, and partially overlapping gene panels from Xenium, and it consistently outperforms prior single-cell expression prediction methods on seven metrics.

Figures

Figures reproduced from arXiv: 2608.08366 by the authors.

Figure 1
Figure 1. Overview of VOICE. VOICE predicts a single cell’s gene expression from its H&E appearance in three stages. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Cell–patch overlap-weighted token pooling for cell [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. In-slide comparison. (a–c) Spatial heatmap of specific gene expressions between ground truth and predicted values [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Cross-slide comparison. (a–c) Spatial heatmap of specific gene expressions between ground truth and predicted values [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Fusion-weight workflow. C Cross-slide analyses C.1 Notation for the cross-tissue gene comparison [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 22 canonical work pages

  1. [1]

    Loïc Chadoutaud, Marvin Lerousseau, Daniel Herrero-Saboya, Julian Ostermaier, Jacqueline Fontugne, Emmanuel Barillot, and Thomas Walter. 2026. sCellST predicts single-cell gene expression from H&E images.Nature Communications 17 (2026), 1194. doi:10.1038/s41467-025-67965-1

    work page doi:10.1038/s41467-025-67965-1 2026

  2. [2]

    Richard J Chen, Tong Ding, Ming Y Lu, Drew FK Williamson, Guillaume Jaume, Andrew H Song, Bowen Chen, Andrew Zhang, Daniel Shao, Muhammad Shaban, et al. 2024. Towards a general-purpose foundation model for computational pathology.Nature medicine30, 3 (2024), 850–862

    2024

  3. [3]

    Tran, Yiwei Xiao, Shengyu Li, Vrutant V

    Weiqing Chen, Pengzhi Zhang, Tu N. Tran, Yiwei Xiao, Shengyu Li, Vrutant V. Shah, Hao Cheng, Kristopher W. Brannan, Keith Youker, Li Lai, Longhou Fang, Yu Yang, Nhat-Tu Le, Jun-ichi Abe, Shu-Hsia Chen, Qin Ma, Ken Chen, Qianqian Song, John P. Cooke, and Guangyu Wang. 2025. A visual–omics foundation model to bridge histopathology with spatial transcriptomi...

    work page doi:10.1038/s41592-025-02707-1 2025

  4. [4]

    Xinlei Chen, Saining Xie, and Kaiming He. 2021. An Empirical Study of Training Self-Supervised Vision Transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 9620–9629. doi:10.1109/ICCV48922.2021.0 0950

    arXiv 2021

  5. [5]

    Nicolas Chiaruttini, Olivier Burri, Peter Haub, Romain Guiet, Jessica Sordet- Dessimoz, and Arne Seitz. 2022. An Open-Source Whole Slide Image Registration Workflow at Cellular Precision Using Fiji, QuPath and Elastix.Frontiers in Com- puter Science3 (2022), 780026. doi:10.3389/fcomp.2021.780026

    arXiv 2022

  6. [6]

    Wei Chow, Linfeng Li, Xian Sun, Lingdong Kong, Zefeng Li, Qi Xu, Hang Song, Tian Ye, Xian Wang, Jinbin Bai, Shilin Xu, Xiangtai Li, Junting Pan, Shaoteng Liu, Ran Zhou, Tianshu Yang, and Songhua Liu. 2026. Masked Generative Transformer Is What You Need for Image Editing. arXiv:2605.10859 [cs.CV] https://arxiv.org/ abs/2605.10859 CVPR Workshop on HiGen

    arXiv 2026

  7. [7]

    Haotian Cui, Chloe Wang, Hassaan Maan, Kuan Pang, Fengning Luo, Nan Duan, and Bo Wang. 2024. scGPT: toward building a foundation model for single-cell multi-omics using generative AI.Nature methods21, 8 (2024), 1470–1480

    2024

  8. [8]

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Imagenet: A large-scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition. Ieee, 248–255

    2009

Show all 35 references
  1. [9]

    Guangyuan Dong, Ziwei Hong, Shenghao Liu, Chenyu Wu, Yuanyuan Fang, Zihao Li, Xudong Zhang, Bingchen Liu, Yuchen Zhang, Haitao Ding, et al. 2026. Dynamic interaction-aware and causality-disentangled framework for multi- modal sentiment analysis.arXiv preprint arXiv:2605.30994(2026)

    2026 arXiv

  2. [10]

    Xiaohang Fu, Yue Cao, Beilei Bian, Chuhan Wang, Dinny Graham, Nirmala Pathmanathan, Ellis Patrick, Jinman Kim, and Jean Yee Hwa Yang. 2025. Spatial gene expression at single-cell resolution from histology using deep learning with GHIST.Nature Methods22 (2025), 1900–1910. doi:1...

    2025 doi

  3. [11]

    Minsheng Hao, Jing Gong, Xin Zeng, Chiming Liu, Yucheng Guo, Xingyi Cheng, Taifeng Wang, Jianzhu Ma, Xuegong Zhang, and Le Song. 2024. Large-scale foundation model on single-cell transcriptomics.Nature methods21, 8 (2024), 1481–1491

    2024

  4. [12]

    Shanshan He, Ruchir Bhatt, Carl Brown, Emily A Brown, Derek L Buhr, Kan Chantranuvatana, Patrick Danaher, Dwayne Dunaway, Ryan G Garrison, Gary Geiss, et al. 2022. High-plex imaging of RNA and proteins at subcellular resolution in fixed tissue by spatial molecular imaging.Natu...

    2022

  5. [13]

    Amanda Janesick, Robert Shelansky, Andrew D Gottscho, Florian Wagner, Stephen R Williams, Morgane Rouault, Ghezal Beliakoff, Carolyn A Morrison, Michelli F Oliveira, Jordan T Sicherman, et al. 2023. High resolution mapping of the tumor microenvironment using integrated single-...

    2023

  6. [14]

    Hongyu Ke, Jack Morris, Yongkang Liu, Satoshi Kitai, Kentaro Oguchi, Yi Ding, and Haoxin Wang. 2026. Deformba: Vision State Space Model with Adaptive State Fusion.arXiv preprint arXiv:2605.21308(2026)

    2026 arXiv

  7. [15]

    Hongyu Ke, Jack Morris, Kentaro Oguchi, Xiaofei Cao, Yongkang Liu, Haoxin Wang, and Yi Ding. 2025. Mambev: Enabling state space models to learn birds- eye-view representations. InThe Thirteenth International Conference on Learning Representations

    2025

  8. [16]

    Cottereau, Changxin Gao, Liang Pan, Wei Tsang Ooi, and Ziwei Liu

    Lingdong Kong, Ao Liang, Tianyi Yan, Hongsi Liu, Wesley Yang, Ziqi Huang, Xian Sun, Wei Yin, Jialong Zuo, Yixuan Hu, Dekai Zhu, Dongyue Lu, Youquan Liu, Guangfeng Jiang, Linfeng Li, Xiangtai Li, Long Zhuo, Lai Xing Ng, Benoit R. Cottereau, Changxin Gao, Liang Pan, Wei Tsang Oo...

    2026 arXiv

  9. [17]

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation. InProceedings of the 39th International Conference on Machine Learn- ing (ICML) (Proceedings of Machine Learning...

    2022

  10. [18]

    Hanzhi Liu, Ningkang Peng, Qianyu Chen, Canran Xiao, et al. 2026. Affordance- first decomposition for continual learning in video-language understanding. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 3908–3919

    2026

  11. [19]

    Koelzer, and Gunnar Rätsch

    Kalin Nonchev, Sebastian Dawo, Karina Silina, Viktor H. Koelzer, and Gunnar Rätsch. 2026. DeepSpot-M: A Multimodal Foundation Model for Transcriptome- Wide Virtual Spatial Transcriptomics from Histology.medRxiv(2026). doi:10.648 98/2026.06.19.26356060

    2026

  12. [20]

    Kalin Nonchev, Glib Manaiev, Viktor H Koelzer, and Gunnar Rätsch. 2025. DeepSpot2Cell: Predicting Virtual Single-Cell Spatial Transcriptomics from H&E images using Spot-Level Supervision.bioRxiv(2025), 2025–09

    2025

  13. [21]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. InInternational conference on machine learnin...

    2021

  14. [22]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. InProceedings ...

    2021

  15. [23]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. 2022. High-resolution image synthesis with latent diffusion models. In 2022 IEEE/CVF conference on computer vision and pattern recognition (CVPR). ieee, 10674–10685

    2022

  16. [24]

    Christina V Theodoris, Ling Xiao, Anant Chopra, Mark D Chaffin, Zeina R Al Sayed, Matthew C Hill, Helene Mantineo, Elizabeth M Brydon, Zexian Zeng, X Shirley Liu, et al . 2023. Transfer learning enables predictions in network biology.Nature618, 7965 (2023), 616–624

    2023

  17. [25]

    Xiyue Wang, Sen Yang, Jun Zhang, Minghui Wang, Jing Zhang, Wei Yang, Jun- zhou Huang, and Xiao Han. 2022. Transformer-based unsupervised contrastive learning for histopathological image classification.Medical image analysis81 (2022), 102559

    2022

  18. [26]

    Jinxi Xiang, Siyu Hou, Yuchen Li, Ryan Quinton, Xiaoming Zhang, Feyisope Eweje, Xiangde Luo, Yijiang Chen, Zhe Li, Colin Bergstrom, Ted Kim, Sierra Willens, Francesca Maria Olguin, Matthew Abikenari, Andrew Heider, Sanjeeth Rajaram, Joel Neal, Maximilian Diehn, Xiang Zhou, and...

    2026 doi arXiv

  19. [27]

    Canran Xiao, Tianxiang Xu, Siyuan Ma, Yiyang Jiang, Haoyu Gao, and Yuhan Wu

  20. [28]

    Chung, Catia T

    Ronald Xie, Kuan Pang, Sai W. Chung, Catia T. Perciani, Sonya A. MacParland, Bo Wang, and Gary D. Bader. 2023. Spatially Resolved Gene Expression Prediction from Histology Images via Bi-modal Contrastive Learning. InAdvances in Neural Information Processing Systems 36 (NeurIPS)

    2023

  21. [29]

    Daiwei Zhang, Amelia Schroeder, Hanying Yan, Haochen Yang, Jian Hu, Michelle Y. Y. Lee, Kyung S. Cho, Katalin Susztak, George X. Xu, Michael D. Feldman, Edward B. Lee, Emma E. Furth, Linghua Wang, and Mingyao Li. 2024. Inferring super-resolution tissue architecture by integrat...

    2024 doi

  22. [30]

    Yabin Zhang, Wenjie Zhu, Hui Tang, Zhiyuan Ma, Kaiyang Zhou, and Lei Zhang

  23. [31]

    Gengmo Zhou, Zhifeng Gao, Qiankun Ding, Hang Zheng, Hongteng Xu, Zhewei Wei, Linfeng Zhang, and Guolin Ke. 2023. Uni-mol: A universal 3d molecular representation learning framework. InThe eleventh international conference on learning representations

    2023

  24. [32]

    Heng Zhou, Jing Tang, Yanshu Li, Canran Xiao, Liwei Hou, Zong Ke, Jiawei Yao, et al. 2026. Comem: Compositional concept-graph memory for vision–language adaptation. InInternational Conference on Learning Representations, Vol. 2026. 20009–20032

    2026

  25. [33]

    Wenjie Zhu, Yabin Zhang, Xin Jin, Wenjun Zeng, and Lei Zhang. 2026. Ants: Adaptive negative textual space shaping for ood detection via test-time mllm understanding and reasoning. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. 20–30. A I...

    2026

  26. [2024]

    In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Dual memory networks: A versatile adaptation approach for vision- language models. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 28718–28728

  27. [2026]

    InInternational Conference on Learning Representations, Vol

    Reversible primitive–composition alignment for continual vision–language learning. InInternational Conference on Learning Representations, Vol. 2026. 88700– 88722

    2026

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.