# arXiv cs.AI: From Cross-Section to Cross-Omics: Conditional Cell Nucleus Image Editing in Spatial Transcriptomics [Canonical reading](https://escapemeridian.com/readings/arxiv-cs-ai-from-cross-section-to-cross-omics-conditional-cell-0f5e4222) The paper’s authors argue that tissue sectioning can make gene expression and cell-image features appear correlated for a measurement-related reason: cutting a three-dimensional tissue at different angles changes the cross-section’s area. Larger cross-sections expose more cytoplasm, which yields higher mRNA counts, while pathomics features such as nuclear area come from the same two-dimensional boundary. The authors apply Manifold-Aware Diffusion (MAD) to edit one feature group while preserving the other. They identify this as addressing a gap in existing gene-to-image methods, which generate images from gene features but do not conditionally edit existing images across both omics layers. In tests on podocyte cells from CosMx SMI kidney tissue, they report that MAD outperforms compared baselines for pathomics editing, with MAE=0.6636 and FID=25.19. A controlled simulation study, they add, finds that MAD’s VAE component preserves cross-group independence better than CCDM, the version without VAE. The results describe performance in the reported evaluations; the abstract does not establish how well the method would work on other tissues or settings. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv cs.AI · Article Original source date: Oct 30, 2026 (2026-10-30). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-10T08:23:56.870Z. First observed: 2026-11-03T09:04:06.810Z. Retrieved: 2026-11-04T01:46:58.484Z. - [From Cross-Section to Cross-Omics: Conditional Cell Nucleus Image Editing in Spatial Transcriptomics](https://arxiv.org/abs/2611.00226) ## Limits The reported evaluation is on podocyte cells from CosMx SMI kidney tissue; the abstract does not report results for other tissues or settings. The evidence provided is the arXiv abstract, which does not describe the evaluation methods or define MAE and FID. ## Open question How would MAD’s editing performance and preservation of cross-group independence compare in evaluations beyond the podocyte cells and controlled simulation described in the abstract? ## Source-stated The authors attribute an apparent relationship between gene expression and pathomics features to tissue-sectioning geometry and measurement, rather than independent biology. > This sectioning geometry creates an apparent correlation between gene expression and pathomics that reflects the measurement process rather than independent biology. - [Supporting source](https://arxiv.org/abs/2611.00226) ## Source-stated The authors apply MAD to edit one feature group while preserving the other, addressing a limitation they identify in existing gene-to-image methods. > enabling targeted modification of one feature group while preserving the other. - [Supporting source](https://arxiv.org/abs/2611.00226) ## Source-stated For pathomics editing on podocyte cells from CosMx SMI kidney tissue, the authors report that MAD outperforms all compared baselines, with the stated MAE and FID values. > Evaluated on podocyte cells from CosMx SMI kidney tissue, MAD achieves MAE=0.6636 and FID=25.19 for pathomics editing, outperforming all compared baselines. - [Supporting source](https://arxiv.org/abs/2611.00226) ## Source-stated In a controlled simulation study, the authors report that MAD’s VAE component preserves cross-group independence better than CCDM. > A controlled simulation study further confirms that the VAE component in MAD better preserves cross-group independence than CCDM (MAD without VAE). - [Supporting source](https://arxiv.org/abs/2611.00226) ## Source and contributor context Yang, Yuechen, Zhu, Junchao, Guo, Junlin, Zhu, Yanfan, Deng, Ruining, Lu, Zhengyi, Yao, Tianyuan, Qu, Chongyu, Lionts, Marilyn, Guo, Xingyi, Wang, Yaohong, Wang, Yu, Zhao, Shilin, Yang, Haichun, Huo, Yuankai · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: f82f2b97-f199-4882-8ebb-7fa4529c5a86. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T08:23:56.870Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 922aa4a3352a0a2a62b160a7d684ff287df271bd9eb22b9295add31fe1eb7c1c. Source text hash: f176373452ca3ee4e235972dd77256f009aeff94a1ec718aced962375e92f64c. ## Revision and source-check history ### Version 1 — 2026-11-10T08:23:56.870Z The paper’s authors argue that tissue sectioning can make gene expression and cell-image features appear correlated for a measurement-related reason: cutting a three-dimensional tissue at different angles changes the cross-section’s area. Larger cross-sections expose more cytoplasm, which yields higher mRNA counts, while pathomics features such as nuclear area come from the same two-dimensional boundary. The authors apply Manifold-Aware Diffusion (MAD) to edit one feature group while preserving the other. They identify this as addressing a gap in existing gene-to-image methods, which generate images from gene features but do not conditionally edit existing images across both omics layers. In tests on podocyte cells from CosMx SMI kidney tissue, they report that MAD outperforms compared baselines for pathomics editing, with MAE=0.6636 and FID=25.19. A controlled simulation study, they add, finds that MAD’s VAE component preserves cross-group independence better than CCDM, the version without VAE. The results describe performance in the reported evaluations; the abstract does not establish how well the method would work on other tissues or settings. ## Original source arXiv cs.AI · Article Original source date: Oct 30, 2026 (2026-10-30). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-10T08:23:56.870Z. First observed: 2026-11-03T09:04:06.810Z. Retrieved: 2026-11-04T01:46:58.484Z. - [From Cross-Section to Cross-Omics: Conditional Cell Nucleus Image Editing in Spatial Transcriptomics](https://arxiv.org/abs/2611.00226) ## Limits The reported evaluation is on podocyte cells from CosMx SMI kidney tissue; the abstract does not report results for other tissues or settings. The evidence provided is the arXiv abstract, which does not describe the evaluation methods or define MAE and FID. ## Open question How would MAD’s editing performance and preservation of cross-group independence compare in evaluations beyond the podocyte cells and controlled simulation described in the abstract? ## Source-stated The authors attribute an apparent relationship between gene expression and pathomics features to tissue-sectioning geometry and measurement, rather than independent biology. > This sectioning geometry creates an apparent correlation between gene expression and pathomics that reflects the measurement process rather than independent biology. - [Supporting source](https://arxiv.org/abs/2611.00226) ## Source-stated The authors apply MAD to edit one feature group while preserving the other, addressing a limitation they identify in existing gene-to-image methods. > enabling targeted modification of one feature group while preserving the other. - [Supporting source](https://arxiv.org/abs/2611.00226) ## Source-stated For pathomics editing on podocyte cells from CosMx SMI kidney tissue, the authors report that MAD outperforms all compared baselines, with the stated MAE and FID values. > Evaluated on podocyte cells from CosMx SMI kidney tissue, MAD achieves MAE=0.6636 and FID=25.19 for pathomics editing, outperforming all compared baselines. - [Supporting source](https://arxiv.org/abs/2611.00226) ## Source-stated In a controlled simulation study, the authors report that MAD’s VAE component preserves cross-group independence better than CCDM. > A controlled simulation study further confirms that the VAE component in MAD better preserves cross-group independence than CCDM (MAD without VAE). - [Supporting source](https://arxiv.org/abs/2611.00226) ## Source and contributor context Yang, Yuechen, Zhu, Junchao, Guo, Junlin, Zhu, Yanfan, Deng, Ruining, Lu, Zhengyi, Yao, Tianyuan, Qu, Chongyu, Lionts, Marilyn, Guo, Xingyi, Wang, Yaohong, Wang, Yu, Zhao, Shilin, Yang, Haichun, Huo, Yuankai · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: f82f2b97-f199-4882-8ebb-7fa4529c5a86. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T08:23:56.870Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 922aa4a3352a0a2a62b160a7d684ff287df271bd9eb22b9295add31fe1eb7c1c. Source text hash: f176373452ca3ee4e235972dd77256f009aeff94a1ec718aced962375e92f64c. ### uncertain — 2026-11-10T08:23:56.870Z — anthropic/claude-opus-5.5 Publication: f82f2b97-f199-4882-8ebb-7fa4529c5a86 attribution — pass: The reading correctly attributes the sectioning-confound argument, the MAD application and all reported results to the paper's authors on arXiv cs.AI. The metrics and the simulation finding are framed as author reports, not established facts. > Authors:Yuechen Yang, Junchao Zhu, Junlin Guo, Yanfan Zhu, Ruining Deng summary — uncertain: The summary is faithful but hard to follow for non-specialists. It never says what spatial transcriptomics is: platforms that measure gene expression and capture nucleus images of the same cells. It also leaves central terms unexplained: pathomics, omics layers, mRNA, podocyte, CosMx SMI, MAE, FID, VAE and CCDM. It does not link why editing one feature group while preserving the other relates to the confound. Smallest repair: add the source's opening sentence and suggest term-library tooltips for these terms. The question is substantive and non-leading. > Spatial transcriptomics (ST) platforms simultaneously measure gene expression and capture cell nucleus images at single-cell resolution. date — pass: The originalPublished date of 2026-10-30 matches the v1 arXiv submission date. It is not the later feed date of 2026-11-03. > \[Submitted on 30 Oct 2026\] support — pass: All four claims match verbatim abstract text. The limitations correctly note the single-tissue evaluation and the missing method and metric definitions. The draft implies no unsupported benefits, costs or real-world reliability, and it keeps the simulation result as the authors' report. > MAD achieves MAE=0.6636 and FID=25.19 for pathomics editing, outperforming all compared baselines. media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the most direct way to access the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/0f5e42220eb5d29bb34c816c92787c2989441cd966776cc3d84199b6b418cfbe/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/arxiv-cs-ai-from-cross-section-to-cross-omics-conditional-cell-0f5e4222.json)