Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Spatial proteomics maps protein distributions in tissues, providing transformative insights for life sciences. However, current sequencing-based technologies suffer from low spatial resolution, and substantial inter-tissue variability in protein expression further compromises the performance of existing molecular data prediction methods. In this work, we introduce the novel task of spatial super-resolution for sequencing-based spatial proteomics (seq-SP) and, to the best of our knowledge, propose the first deep learning model for this task—Neural Proteomics Fields (NPF). NPF formulates seq-SP as a protein reconstruction problem in continuous space by training a dedicated network for each tissue. The model comprises a Spatial Modeling Module, which learns tissue-specific protein spatial distributions, and a Morphology Modeling Module, which extracts tissue-specific morphological features. Furthermore, to facilitate rigorous evaluation, we establish an open-source benchmark dataset, Pseudo-Visium SP, for this task. Experimental results demonstrate that NPF achieves state-of-the-art performance with fewer learnable parameters, underscoring its potential for advancing spatial proteomics research. Our code and dataset are publicly available at https://github.com/Bokai-Zhao/NPF.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1305_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/Bokai-Zhao/NPF

Link to the Dataset(s)

Human Tonsil with Add-on Antibodies: https://www.10xgenomics.com/datasets/visium-cytassist-gene-and-protein-expression-library-of-human-tonsil-with-add-on-antibodies-h-e-6-5-mm-ffpe-2-standard Human Tonsil: https://www.10xgenomics.com/datasets/gene-protein-expression-library-of-human-tonsil-cytassist-ffpe-2-standard Glioma spatialomics dataset: https://zenodo.org/records/12624860

BibTex

@InProceedings{ZhaBok_Neural_MICCAI2025,
        author = { Zhao, Bokai and Shi, Weiyang and Chao, Hanqing and Yang, Zijiang and Zhang, Yiyang and Song, Ming and Jiang, Tianzi},
        title = { { Neural Proteomics Fields for Super-resolved Spatial Proteomics Prediction } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15967},
        month = {September},
        page = {384 -- 394}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper aims at tackling the challenge of spatial proteomics “super resolution”. The authors propose a multi-stream network, composed of a dedicated spatial modeling stream focusing on coordinates and an image encoding stream focusing on WSI patch embedding. Experiments on two datasets demonstrate the effectiveness of the proposed method.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    1. The topic is timely and relevant to the community, aligning with the growing interest in spatial omics and super-resolution techniques.

    2. The overall structure of the paper is clear and well-organized.

    3. Although somewhat limited in quantitative scope, the experimental design is thoughtful and provides supportive evidence for the method’s effectiveness.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    1. The methodological novelty is somewhat limited. Most components are adaptations or combinations of ideas from prior works, including NeRF, UNI, and adapter-based networks. As such, the approach may not offer substantial new insights to the field.

    2. The use of a dedicated stream for processing spatial coordinates seems unusual to me. While the ablation studies support its effectiveness, it would be valuable to know whether early fusion strategies—such as integrating positional information into the image encoder at an early stage—were explored or compared.

    3. The virtual sampling strategy used in constructing the Pseudo-Visium SP dataset requires further clarification. Since the dataset is used for model training, it is essential to demonstrate that the ground truth accurately reflects real-world protein distributions. The authors should better explain how virtual augmentation preserves biological validity.

    4. The definition of the “super-resolution” task in spatial omics remains ambiguous. Many existing methods frame it as a regression task that predicts spot-level expression based on WSI patches. From a traditional computer vision perspective, true super-resolution would involve predicting gene or protein expression within smaller or denser spatial units. While the community has largely adopted the current naming convention, I suggest the authors consider renaming the paper to “Neural Proteomics Fields for Super-resolved Spatial Proteomics Prediction” for greater clarity.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    I would very much like to see the authors’ rebuttal with respect to the raised issues from the reviewers. I would be happy to raise my score if the rebuttal satisfactorily resolves my questions.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (3) Weak Reject — could be rejected, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    This paper addresses an interesting and timely topic in bioengineering, and the proposed method appears to be feasible. However, the lack of clarity regarding dataset preprocessing and the absence of comparisons with other positional encoding methods weaken the overall reliability of the work.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    The rebuttal demonstrates a positive and constructive attitude and addresses most of my concerns satisfactorily. Although the explanation regarding the novelty of the work remains somewhat unconvincing, I believe the overall contribution has merit. Therefore, I have raised my rating from weak reject to weak accept, assuming the authors are willing to release their code, dataset, and/or visualization results to support reproducibility and further validation.



Review #2

  • Please describe the contribution of the paper

    This paper proposes Neural Proteomics Fields (NPF) to achieve sequencing-based spatial proteomics super-resolution. The Spatial Modeling Module encodes spatial information in a continuous manner, while the Morphology Modeling Module captures tissue-specific features using a Pathology-Foundation Model. By integrating these two branches, NPF achieves superior performance on the proposed Spatial Proteomics Prediction Benchmark Dataset.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
    1. The introduction of position encoding from NeRF view is interesting.
    2. The Morphology Modeling Module combine multi-scale feature and foundation model’s prior is novel. 3.The main experiments are comprehensive, clearly demonstrating the framework’s superiority over competing methods. 4.The paper is well-written, logically structured, and easy to follow.
  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
    1. The proposed framework bears similarities to [1], as both utilize 2D coordinates, pathology images, and UNI-based features as inputs. However, the paper lacks a discussion or direct comparison with this prior work.
    2. The ablation study settings are unclear. For instance, does the fifth column in Table 4 represent results using only coordinates as input without the Spatial Modeling Module (SMM)? Similarly, does the sixth column reflect results using pyramid convolutional network features without the Tissue-Specific Feature Encoder (TSFE)? Clarification is needed.
    3. Given the dual-branch architecture’s similarity to [1], additional experiments are necessary to demonstrate the effectiveness of the TSFE compared to simply concatenating UNI features with H&E based multiscale features. [1] Shi, Zhiceng, et al. “High-resolution spatial transcriptomics from histology images using histosge.” 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). IEEE, 2024.
  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (3) Weak Reject — could be rejected, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Although the paper presents a clear and interesting method, the lack of comparison with similar approach leads me to recommend this score. I would be happy to raise the score if my concerns are addressed.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    Thanks for the author’s response, which effectively addressed my concerns. Given the novel perspective in position encoding and the comprehensive experimental evidence, I raise my rating.



Review #3

  • Please describe the contribution of the paper

    This paper introduces a novel task termed “spatial proteomics (seq-SP) super-resolution reconstruction” and proposes a dedicated deep learning framework called Neural Proteomics Fields (NPF). The model integrates spatial coordinate encoding with tissue morphological feature modeling, featuring a dual-branch architecture comprising a spatial modeling module and a morphological modeling module to achieve high-resolution reconstruction of spatial protein expression.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    This paper introduces a novel task termed “spatial proteomics (seq-SP) super-resolution reconstruction” and proposes a dedicated deep learning framework called Neural Proteomics Fields (NPF). The model integrates spatial coordinate encoding with tissue morphological feature modeling, featuring a dual-branch architecture comprising a spatial modeling module and a morphological modeling module to achieve high-resolution reconstruction of spatial protein expression.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    Below is my detailed evaluation of this work:

    1. The study focuses on the cutting-edge challenge of seq-SP super-resolution reconstruction, which has yet to be systematically explored, demonstrating exploratory significance. However, the introduction inadequately emphasizes the unique importance of protein-level research and its biological implications. The authors are advised to further highlight these aspects.

    2. While the paper proposes a deep learning-based seq-SP super-resolution task and designs the corresponding NPF model, its innovation lacks in-depth analysis and critique of existing work. Several related deep learning methods (e.g., STNet and istar in spatial transcriptomics) have been reported in the literature. Although these primarily target transcriptomic data, their core ideas and methodological frameworks significantly overlap with the approach presented here.

    3. The authors have constructed the first open-source seq-SP super-resolution dataset, Pseudo-Visium SP, providing a benchmark resource for future research. The experimental results demonstrate the superiority of NPF. However, additional visualizations on actual tissue sections could further enhance the intuitive understanding of the model’s predictive performance.

    4. The comparative methods in this study are primarily drawn from the spatial transcriptomics domain, lacking comparisons with representative methods in proteomics or other image reconstruction fields. This may fail to comprehensively reflect the model’s actual advantages.

    5. The authors could incorporate downstream analyses, such as the biological significance enabled by this method, to showcase the model’s practical utility in biomedical research.

  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not provide sufficient information for reproducibility.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (4) Weak Accept — could be accepted, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    This paper introduces a novel task termed “spatial proteomics (seq-SP) super-resolution reconstruction” and proposes a dedicated deep learning framework called Neural Proteomics Fields (NPF). The model integrates spatial coordinate encoding with tissue morphological feature modeling, featuring a dual-branch architecture comprising a spatial modeling module and a morphological modeling module to achieve high-resolution reconstruction of spatial protein expression.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    After carefully considering the authors’ rebuttal and clarifications, I have decided to raise my initial evaluation from Weak Accept to Accept. The authors have adequately addressed the main concerns raised in the initial review.



Author Feedback

We sincerely thank all reviewers for their thoughtful and constructive feedback. We appreciate the recognition of the timeliness and novelty of our task (R1, R2), the interesting nature of our approach (R3), and the comprehensiveness and soundness of our experiments (R2, R3). Below, we provide point-to-point responses to the primary concerns raised by the reviewers.

  1. Key Innovations (R1.2, R2.1, R3.1): Our main innovation lies in reformulating the task of super-resolved spatial omics prediction as a continuous protein spatial distribution reconstruction problem, rather than treating it as an isolated spot-level regression problem as in prior works. This formulation allows our model to leverage coordinate information in a fundamentally different way, which we elaborate on in the next response. Furthermore, the role of histological image information remains critical. Representative methods such as istar and STNet either rely solely on pretrained models, lackin flexibility in modeling tissue-specific protein distributions, or train from scratch, leading to reduced robustness in noisy conditions with limited training data. In contrast, our framework combines the TSFE with the pretrained UNI, achieving a desirable balance between customization and generalization robustness.

  2. Spatial Modeling Module (SMM) (R2.2, R3.1): The core purpose of our SMM is to enable continuous mapping from arbitrary spatial coordinates to protein expression values by learning the latent spatial distribution of proteins. Achieving this objective requires a dedicated and sufficiently deep standalone module, rather than a simple concatenation of coordinate and image features. As such, simple early fusion of coordinates and image features (e.g., as in [1] suggested by R3) is inadequate. The 2-dimensional coordinates tend to be overwhelmed by high-dimensional image features, making it difficult for the model to capture meaningful spatial relationships. Our ablation results in Tables 3 and 4 already validate the effectiveness of SMM. Additionally, we quantitatively compared SMM with early fusion on our benchmarks, and found that early fusion achieves MSE = 0.274 and PCC = 0.8391, markedly inferior to SMM, and not significantly better than using image features alone. In future work, we aim to further enhance the SMM to support a broader range of spatial omics tasks involving continuous spatial inference.

  3. Dataset Construction and Validity (R1.3, R2.3): Our virtual sampling approach provides a principled and practical framework for simulating low-resolution spatial omics data from high-resolution sources. Similar strategies have been adopted in existing studies [12]. Specifically, we simulate lower-resolution sampling by integrating protein intensities over circular ROIs and applying a configurable sampling rate to mimic capture efficiency. To validate the biological plausibility of our simulation, we visualized the distributions of key proteins such as CD4, and confirmed that the simulated spatial patterns align closely with real data. All simulation details and visualizations will be released with the dataset code upon acceptance to support reproducibility and transparency.

  4. Responses to Other Comments: R2.4: Thank you for the helpful suggestion. We will adopt the term “super-resolved” in our final camera-ready version. R3.2: The checkmarks (√) in Table 4 denote the modules used. When SMM is not employed, coordinate inputs are omitted, and only image patches are used. A comparison between the 6th and 7th columns shows the contribution of TSFE to the overall performance.



Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.

    N/A

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    This paper addresses the novel task of spatial proteomics super‐resolution by introducing Neural Proteomics Fields, a dual‐branch framework that combines continuous coordinate encoding with tissue‐specific morphology modeling, and provides the first open‐source Pseudo‐Visium SP benchmark dataset. All three reviewers raised their scores to “Accept” in their post‐rebuttal reviews, noting that the authors had addressed major concerns and ensured reproducibility. Based on the post‐rebuttal comments and demonstrated performance, I recommend acceptance.



Meta-review #3

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A



back to top