List of Papers Browse by Subject Areas Author List
Abstract
We present GeneMorphFormer, a Transformer-based model to decode nonlinear interactions between gene expression and cortical morphology. We align expression maps with gray matter and white matter boundary curves through spatial normalization by leveraging marmoset in situ hybridization (ISH) data. Our model employs multi-head self-attention to model global dependencies across 1024 gene features, optimized by a hybrid loss (MSE and Hausdorff distance) balancing local precision and global shape fidelity. SHapley Additive exPlanations (SHAP) analysis is subsequently employed to quantify the contribution of genes to morphological shape. Wavelet-based clustering further reveals distinct gene sets governing smooth versus fluctuating morphologies, suggesting hierarchical genetic regulation. Experimental results demonstrate that GeneMorphFormer outperforms traditional networks in both global shape matching and local precision. This work proposed a biologically interpretable Transformer architecture for cross-scale gene-morphology mapping and enables systematic exploration of genetic drivers in cortical morphology malformations.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1853_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Leveup/GeneMorphFormer
Link to the Dataset(s)
Marmoset ISH dataset: https://gene-atlas.brainminds.jp/
BibTex
@InProceedings{LiXia_GeneMorphFormer_MICCAI2025,
author = { Li, Xiao and Zhang, Han and Sun, Qitai and Jia, Chenjie and He, Xiaowei and Ren, Yudan},
title = { { GeneMorphFormer: Transformer-Driven Cross-Scale Mapping from Gene Expression to Cortical Morphology } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15973},
month = {September},
page = {248 -- 257}
}
Reviews
Review #1
- Please describe the contribution of the paper
I think the main contribution of this paper is the author proposes a transfomer based deep learning model named GeneMorFormer to explore the nonlinear relationship between the gene expression and the cortical morphology malformations.
- 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.
I think the main novelty is the proprosed model, and within the model,the author propose several novel design such as the data processing, the wavelet-based clustering techniques. The final experiment also show the improved performance compared to several traditional networks.
- 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.
The major weakness is the experiment section, in table 1, only three very traditional methods have been compared including CNN, Resnet and GCN. and the metrics to evaluate is MSE, MAE and Hausdorff distance. Overall, the whole experiment lack comprehensenive design and lacks realiability to desmonstrate the performance of the proposed method.
- 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure 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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Although the author adapts the transformer based deep learning techniques to a novel domain in gene experission, the overal design of the method and the extremely limited experiments design lead me to reject this paper.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
The authors use a transformer model, and shapley analysis to correlate gene expression along the cortical boundary of primate brain slices, with the shape of the cortical boundary.
- 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.
The approach is novel, regressing gene expression to 2D cortical boundary curves. Additionally, both their method and their results are fairly clear. Similar genes are identified through shapley explainability in both their curve regression results, and their wavelet clustering results, which helps validate their gene findings.
- 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.
- Given the symmetry of the brain, and that they reverse each curve in their dataset, many of the samples in their test set are near identical reflections of elements in their training set - it would be more meaningful to keep whole brain slices, or a whole hemisphere for testing
- There is very limited discussion of the genes that were found with SHAP analysis - how many of these genes were also found in the works they cited by Li et al. for example? Why don’t they think they recovered Trnp1 or TMEM14B?
- It seems like the code and preprocessed data for this work could be easily released, why don’t the authors do this?
Minor:
- Figure 4 does not reference the subplots, limiting clarity
- To me, the obvious partition of the curves in the wavelet clustering is the y-coordinate at x=300. Decribing the sets as “smooth” and “high-variance” is not very exact.
- 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 submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure 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.
(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?
As it stands, their main contribution is description of a method. I think there are some limitations of a) the rigor of their train/test split b) the meaning of the genes they identified and c) the reproducibility. I think if the authors work to mitigate these concerns (alternative train/test split, more discussion on genes found, code/data release), then this paper would be worthy of acceptance.
- 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.
The authors adequately addressed my initial concerns.
Review #3
- Please describe the contribution of the paper
The main contribution of the paper is the development and demonstration of GeneMorphFormer, a novel Transformer-based framework that directly links gene expression profiles to cortical morphology. The model leverages multi-head self-attention to capture the complex and long-range dependencies among 1024 gene expression features. This model is designed to decode the non linear interactions that underline the shaping of cortical boundaries. Traditional methods used simple concatenation which was in effective in capturing the relationships between genes and brain shape. GeneMorphFormer is tailored for predicting the 2D spatial coordinates of cortical boundary curves. This enables the model to accurately capture both local and global morphological characteristics of the cortex.
- 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.
- The paper is well organized. The figures are neat and illustrative.
- The introduction of GeneMorphFormer, a Transformer-based framework, represents a significant methodological advancement. Unlike traditional models that rely on simple feature concatenation, this architecture leverages multi-head self-attention to capture long-range, nonlinear dependencies between 1024 gene expression features.
- The Paper provides interpretable model insights via SHAP analysis into which genes drive cortical morphology.
- The extensive evaluation, including both quantitative metrics (lower MSE and Hausdorff distance compared to CNNs, ResNets, and GCNs) and visual comparisons, highlights the practical benefits of the approach.
- 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.
-
There is a lack of detailed ablation studies that isolate the impact of key components, such as the contribution of spatial normalization, the hybrid loss function, or wavelet-based clustering. Without these studies, it is challenging to determine which parts of the pipeline most significantly contribute to the observed performance improvements.
-
The paper compares with standard CNNs, ResNets, GCNs, but lacks comparison with other recent transformer-based approaches in genomics.
-
- 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 submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure 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.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper has good representation of the problem, solution, and results.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
We sincerely thank all reviewers for their constructive and insightful feedback. We appreciate the thoughtful evaluation and respond to the comments point-by-point below. Common Concern: Reproducibility of the paper (R1, R2, R3): Due to double-blind review requirements, we have not released code or data during the review. Upon acceptance, we will release the full code and dataset to ensure reproducibility. Response to Reviewer #1:
- Data Splitting and Flipped Curves: We use flipping of cortical curves as a data augmentation strategy to improve generalization and reduce overfitting. We acknowledge the concern regarding potential information leakage. During method development, we explored different data partitioning strategies, including those that withheld an entire slice for testing or omitted curve flipping altogether. These strategies yielded generally consistent results, though with slight variations in accuracy. Due to space limitations, we reported the most effective setup in the paper.
- Identified Genes and Absence of Trnp1/TMEM14B: Trnp1 and TMEM14B were cited in the Introduction as motivational examples of gene–morphology links. These genes are not yet included in the Brain/MINDS marmoset ISH dataset. We also discovered genes similar to those cited in the literature, such as NFIX, PJA2, MPPED1, HMP19, and NEUROD6. These genes have been shown to be closely related to cortical morphology. Moreover, we identified a group of new genes by this method, providing a new perspective for exploring the potential genetic mechanisms driving cortical folding. Minor 1. Clarification of Figure 4 Subplots: We will revise the manuscript to explicitly reference and label each subplot in Figure 4, and refine the caption and related text for clarity and rigor. Minor 2. Curve Clustering Terminology: To capture global and local gene–morphology patterns, we apply Discrete Wavelet Transform (DWT) for multi-scale feature extraction, followed by K-Means clustering. We acknowledge that the terms “smooth” and “fluctuating” may be imprecise and will revise Section 2.5 to adopt more neutral terms (e.g., “Cluster A” and “Cluster B”). Response to Reviewer #2:
- Experimental Design and Evaluation Metrics: Gene regulation has been suggested as an important factor influencing cortical folding. However, due to the complex relationship between gene expression and cortical morphology, systematically uncovering this link remains a challenging task. Our study aims to explore this relationship by modeling gene–morphology associations based on spatial gene expression data. We believe our approach provides a potentially useful perspective for identifying candidate regulatory genes and contributes toward a better understanding of the genetic basis of cortical structure. Since this is a relatively underexplored area, few comparable deep learning baselines exist, we selected conventional methods for comparison. We will expand our comparison scope and incorporate additional evaluation metrics in future work. Due to page limits, we were unable to include ablation studies on core components. We prioritized presenting the complete workflow and key contributions. We acknowledge that this may have given the impression of limited experimental design and will clarify these components more explicitly in the future version. Response to Reviewer #3:
- Lack of Detailed Ablation Studies: We unanimously agree that ablation studies can elucidate the contributions of key components. However, due to the limitation of article length, we only presented the best output results in this article to ensure that all our work content and innovative performance were fully presented.
- Lack of Comparison with Transformer-based Genomic Methods: Of course, comparing with related Transformer based methods is very valuable. We will continue to monitor the latest developments in the related field and further expand our comparative scope. We thank the reviewers again for their feedback.
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.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
Applying transformers and their variants to genetic data analysis is not a novel concept. The major concerns with this paper lie in the lack of justification for the model design and insufficient validation. Without a thorough analysis and discussion of the rationale behind the chosen architecture, as well as strong empirical evidence demonstrating its advantages over existing transformer-based models in genetic data analysis, it is difficult to assess the true value and contribution of this work.
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