CoChat

The DOI matches its Crossref record; 38 of the 38 references it cites were found in a registry or an index.

MCGAE: unraveling tumor invasion through integrated multimodal spatial transcriptomics

Open access
Yang, Yiwen, Zhang, Chengming, Liu, Zhaonan, Aihara, Kazuyuki, Zhang, Chuanchao, Chen, Luonan, Wei, Wu · 2024 · Briefings in Bioinformatics

Abstract

Spatially Resolved Transcriptomics (SRT) serves as a cornerstone in biomedical research, revealing the heterogeneity of tissue microenvironments. Integrating multimodal data including gene expression, spatial coordinates, and morphological information poses significant challenges for accurate spatial domain identification. Herein, we present the Multi-view Contrastive Graph Autoencoder (MCGAE), a cutting-edge deep computational framework specifically designed for the intricate analysis of spatial transcriptomics (ST) data. MCGAE advances the field by creating multi-view representations from gene expression and spatial adjacency matrices. Utilizing modular modeling, contrastive graph convolutional networks, and attention mechanisms, it generates modality-specific spatial representations and integrates them into a unified embedding. This integration process is further enriched by the inclusion of morphological image features, markedly enhancing the framework’s capability to process multimodal data. Applied to both simulated and real SRT datasets, MCGAE demonstrates superior performance in spatial domain detection, data denoising, trajectory inference, and 3D feature extraction, outperforming existing methods. Specifically, in colorectal cancer liver metastases, MCGAE integrates histological and gene expression data to identify tumor invasion regions and characterize cellular molecular regulation. This breakthrough extends ST analysis and offers new tools for cancer and complex disease research.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1093/bib/bbae608

Cites

38 of 38 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Overview of MCGAE. (A) Spatial transcriptomics techniques generate diverse data modalities, including gene expression profiles, cellular positioning, and morphological images. We process these modalities to create multiple views, denoted X_1,X_2,…,X_v and a_1,a_2,…,a_w. In the case of morphological images, we segment them based on cell coordinates, often H&E stained, to extract relevant features. (B) For each view, we use GCNs to integrate gene expression and spatial coordinates to learn spot representations. Simultaneously, we employ self-supervised contrastive learning on the base graph (G_1^a = G_1^X) to capture genuine biological signals (Z_1). For morphological images, we extract their representations using a ResNet50 network and map them into a spot-like representation space with an MLP. (C) We fuse representations of multiple views invariant to X into Z^X using selfattention, and representations fixed to a into Z^A. These are fused with the Z^morph view of the image into the final representation Z. We optimize the model with unsupervised iterative clustering loss. Additionally, we reconstruct A and X based on the fused representations (Z^A,Z^X). (D) The final representation Z generated by MCGAE, along with the reconstructed expression data, can be applied to various downstream analysis tasks, including spatial domain analysis, trajectory inference, identification of SVGs, data denoising, and the elucidation 3D spatial domains.
Figure 1 Overview of MCGAE. (A) Spatial transcriptomics techniques generate diverse data modalities, including gene expression profiles, cellular positioning, and morphological images. We process these modalities to create multiple views, denoted X_1,X_2,…,X_v and a_1,a_2,…,a_w. In the case of morphological images, we segment them based on cell coordinates, often H&E stained, to extract relevant features. (B) For each view, we use GCNs to integrate gene expression and spatial coordinates to learn spot representations. Simultaneously, we employ self-supervised contrastive learning on the base graph (G_1^a = G_1^X) to capture genuine biological signals (Z_1). For morphological images, we extract their representations using a ResNet50 network and map them into a spot-like representation space with an MLP. (C) We fuse representations of multiple views invariant to X into Z^X using selfattention, and representations fixed to a into Z^A. These are fused with the Z^morph view of the image into the final representation Z. We optimize the model with unsupervised iterative clustering loss. Additionally, we reconstruct A and X based on the fused representations (Z^A,Z^X). (D) The final representation Z generated by MCGAE, along with the reconstructed expression data, can be applied to various downstream analysis tasks, including spatial domain analysis, trajectory inference, identification of SVGs, data denoising, and the elucidation 3D spatial domains.CC BY-NC · PubMed Central
Figure 2 MCGAE has shown improvements in spatial domain recognition on both DLPFC and mouse brain anterior datasets. (A) Boxplot displays the ARI scores of the twelve methods applied to the 12 DLPFC slices. In the boxplot, the central line represents the median score, the box boundaries represent the upper and lower quartiles, and the whiskers extend up to 1.5 times the interquartile range, similar to what is shown in (E) and (H). (B) H&E image and manual annotation from the original study. (C) The clustering results for slice 151,669 of the DLPFC dataset were obtained using MCGAE, GraphST, Spaceflow, ConST, SpaGCN, and STAGATE. Clustering results for other algorithms can be found in Supplementary Fig. S5. (D) Trajectory inference was generated using embeddings from MCGAE, GraphST, Spaceflow, ConST, SpaGCN, and STAGATE for the DLPFC section 151669. (E) Boxplot is used to display the Moran’s I values for SVGs detected by MCGAE, SpaGCN and raw (raw expression), and for slice 151,669. (F) the visualization of denoised and raw expression patterns of layer marker genes. (G) Spatial domain clustering based on the annotation labels. (H) Boxplot displays the ARI scores of the twelve methods applied to the mouse brain anterior dataset. (I) The clustering results of the mouse brain anterior dataset were obtained using MCGAE, BayesSpace, STAGATE, GraphST, ConST, SpaceFlow, Seurat, STMGCN.
Figure 2 MCGAE has shown improvements in spatial domain recognition on both DLPFC and mouse brain anterior datasets. (A) Boxplot displays the ARI scores of the twelve methods applied to the 12 DLPFC slices. In the boxplot, the central line represents the median score, the box boundaries represent the upper and lower quartiles, and the whiskers extend up to 1.5 times the interquartile range, similar to what is shown in (E) and (H). (B) H&E image and manual annotation from the original study. (C) The clustering results for slice 151,669 of the DLPFC dataset were obtained using MCGAE, GraphST, Spaceflow, ConST, SpaGCN, and STAGATE. Clustering results for other algorithms can be found in Supplementary Fig. S5. (D) Trajectory inference was generated using embeddings from MCGAE, GraphST, Spaceflow, ConST, SpaGCN, and STAGATE for the DLPFC section 151669. (E) Boxplot is used to display the Moran’s I values for SVGs detected by MCGAE, SpaGCN and raw (raw expression), and for slice 151,669. (F) the visualization of denoised and raw expression patterns of layer marker genes. (G) Spatial domain clustering based on the annotation labels. (H) Boxplot displays the ARI scores of the twelve methods applied to the mouse brain anterior dataset. (I) The clustering results of the mouse brain anterior dataset were obtained using MCGAE, BayesSpace, STAGATE, GraphST, ConST, SpaceFlow, Seurat, STMGCN.CC BY-NC · PubMed Central
Figure 3 MCGAE accurately identifies spatial domains in the HBC and colorectal cancer datasets using histological images. (A) Morphological images and manual annotations for HBC data. (B) Boxplot displays the ARI scores of the twelve methods applied to the HBC dataset. In the boxplot, the central line represents the median score, the box boundaries represent the upper and lower quartiles, and the whiskers extend up to 1.5 times the interquartile range. (C) Clustering results for the HBC dataset were obtained using MCGAE, MCGAE without morphological information, GraphST, SpaGCN, and STAGATE, with additional algorithm results in Supplementary Fig. S8. (D) The spatial expression patterns of SVGs detected by MCGAE were visualized using both the denoised and original expression data. (E) Colorectal cancer morphological images with an oncology pathologist’ annotation. (F) The clustering results for the colorectal cancer dataset were obtained using MCGAE, MCGAE without morphological information, and STAGATE, with additional algorithm results presented in Supplementary Fig. S9. The additional contour lines in the figure represent the contours of the domains of interest for each method, with annotations consistent with those in (E) (G) Visualization of marker genes for tumor using denoised and original expression data from MCGAE. (H) the GO enrichment plot for SVGs in tumors located at the boundary between tumor and connective tissue depicts biological processes, cellular components, and molecular functions using different shading patterns to distinguish the categories.
Figure 3 MCGAE accurately identifies spatial domains in the HBC and colorectal cancer datasets using histological images. (A) Morphological images and manual annotations for HBC data. (B) Boxplot displays the ARI scores of the twelve methods applied to the HBC dataset. In the boxplot, the central line represents the median score, the box boundaries represent the upper and lower quartiles, and the whiskers extend up to 1.5 times the interquartile range. (C) Clustering results for the HBC dataset were obtained using MCGAE, MCGAE without morphological information, GraphST, SpaGCN, and STAGATE, with additional algorithm results in Supplementary Fig. S8. (D) The spatial expression patterns of SVGs detected by MCGAE were visualized using both the denoised and original expression data. (E) Colorectal cancer morphological images with an oncology pathologist’ annotation. (F) The clustering results for the colorectal cancer dataset were obtained using MCGAE, MCGAE without morphological information, and STAGATE, with additional algorithm results presented in Supplementary Fig. S9. The additional contour lines in the figure represent the contours of the domains of interest for each method, with annotations consistent with those in (E) (G) Visualization of marker genes for tumor using denoised and original expression data from MCGAE. (H) the GO enrichment plot for SVGs in tumors located at the boundary between tumor and connective tissue depicts biological processes, cellular components, and molecular functions using different shading patterns to distinguish the categories.CC BY-NC · PubMed Central
Figure 4 MCGAE facilitates the study of tumor invasiveness in colorectal cancer liver metastasis cases. (A) Colorectal cancer morphological images with an oncology pathologist’s annotation. (B) Clustering results for the colorectal cancer dataset were obtained using MCGAE, GraphST, and STAGATE, with additional algorithm results in Supplementary Fig. S10. (C) A t-SNE plot of all spots in the sample, colored according to manual annotation and region Y. (D) A refined t-SNE plot of region Y, segmented into three distinct categories, highlighting the top DEGs for each category. (E) The three categories of region Y projected back onto the original ISH slides. (F) A GO enrichment plot for DEGs across the three categories in region Y, with colors corresponding to each category. (G) Liver cancer morphological images with an oncology pathologist’s annotation. (H) Clustering results for the liver cancer dataset were obtained using MCGAE, GraphST, and STAGATE, with additional algorithm results in Supplementary Fig. S12. (I) Denoised and original expression patterns of metabolism-related gene CYP3A4 in liver cancer dataset. (J) The heatmap of common marker genes in liver cancer and normal liver tissue. (K) A boxplot showing the expression differences of common SVGs across domains 5, 2, and 6. (L) Enrichment analysis for SVGs in domain 2 relative to domain 5 and 6, along with their shared SVGs.
Figure 4 MCGAE facilitates the study of tumor invasiveness in colorectal cancer liver metastasis cases. (A) Colorectal cancer morphological images with an oncology pathologist’s annotation. (B) Clustering results for the colorectal cancer dataset were obtained using MCGAE, GraphST, and STAGATE, with additional algorithm results in Supplementary Fig. S10. (C) A t-SNE plot of all spots in the sample, colored according to manual annotation and region Y. (D) A refined t-SNE plot of region Y, segmented into three distinct categories, highlighting the top DEGs for each category. (E) The three categories of region Y projected back onto the original ISH slides. (F) A GO enrichment plot for DEGs across the three categories in region Y, with colors corresponding to each category. (G) Liver cancer morphological images with an oncology pathologist’s annotation. (H) Clustering results for the liver cancer dataset were obtained using MCGAE, GraphST, and STAGATE, with additional algorithm results in Supplementary Fig. S12. (I) Denoised and original expression patterns of metabolism-related gene CYP3A4 in liver cancer dataset. (J) The heatmap of common marker genes in liver cancer and normal liver tissue. (K) A boxplot showing the expression differences of common SVGs across domains 5, 2, and 6. (L) Enrichment analysis for SVGs in domain 2 relative to domain 5 and 6, along with their shared SVGs.CC BY-NC · PubMed Central
Figure 5 MCGAE handles datasets from both stereo-seq and slide-seq technologies and can construct 3D spatial domains. (A) Laminar organization of the mouse olfactory bulb annotated using the DAPI-stained image. (B) Spatial domains identified by MCGAE, GraphST and STAGATE in the mouse olfactory bulb Stereoseq data. (C) Visualization of the spatial domains identified by MCGAE and the corresponding marker gene expressions. The identified domains are aligned with the annotated laminar organization of the mouse olfactory bulb. (D) Visualization of the 3D hippocampal volume stacked by seven aligned consecutive sections profiled by slide-seq. (E) 3D spatial domain recognition is achieved by extending MCGAE, where we construct a large adjacency matrix covering all batches and focus on the adjacency distances between adjacent sections. (F) Cluster assignments using conventional adjacency matrix calculation in MCGAE and UMAP plots based on MCGAE embeddings. Spots are color-coded for identified spatial domains and section IDs. (G) Cluster assignments using extended adjacency matrix calculation in MCGAE-3D and UMAP plots based on MCGAE-3D embeddings. Spot colors represent spatial domains and section IDs.
Figure 5 MCGAE handles datasets from both stereo-seq and slide-seq technologies and can construct 3D spatial domains. (A) Laminar organization of the mouse olfactory bulb annotated using the DAPI-stained image. (B) Spatial domains identified by MCGAE, GraphST and STAGATE in the mouse olfactory bulb Stereoseq data. (C) Visualization of the spatial domains identified by MCGAE and the corresponding marker gene expressions. The identified domains are aligned with the annotated laminar organization of the mouse olfactory bulb. (D) Visualization of the 3D hippocampal volume stacked by seven aligned consecutive sections profiled by slide-seq. (E) 3D spatial domain recognition is achieved by extending MCGAE, where we construct a large adjacency matrix covering all batches and focus on the adjacency distances between adjacent sections. (F) Cluster assignments using conventional adjacency matrix calculation in MCGAE and UMAP plots based on MCGAE embeddings. Spots are color-coded for identified spatial domains and section IDs. (G) Cluster assignments using extended adjacency matrix calculation in MCGAE-3D and UMAP plots based on MCGAE-3D embeddings. Spot colors represent spatial domains and section IDs.CC BY-NC · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 36 references

Cited by

See all 6

Metadata

Authors
Yang, Y., Zhang, C., Liu, Z., Aihara, K., Zhang, C., Chen, L., & Wei, W.
Affiliations
  • Lingang Laboratory , Building 8, 319 Yueyang Road, Xuhui District, Shanghai 200031 ,
  • University of Chinese Academy of Sciences (CN)
  • Shanghai Jiao Tong University (CN)
  • The University of Tokyo Institutes for Advanced Study (JP)
  • Chinese Academy of Sciences (CN)
Corresponding author
  • Zhang, Chuanchao
  • Chen, Luonan
  • Wei, Wu
Venue
Briefings in Bioinformatics
Published
2024-11-22
Volume
26
Issue
1
ISSN
  • 1467-5463 (ISSN-L)
  • 1477-4054
OpenAlex
W4404637787
Cited by
5
References
36
Citations by year
2026: 4 · 2025: 1
FWCI
0.68
Type
journal article
Language
en
Open access
Locations
  • Briefings in Bioinformatics (publishedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
Funders
  • JST Moonshot R&D (JPMJMS2021)
  • Major Science and Technology R&D Project of the Science and Technology Department of Jiangxi Province (20213AAG01013])
  • Science and Technology Commission of Shanghai Municipality (23JS1401300)
  • National Natural Science Foundation of China (62202120, 12131020, T2341007, T2350003, 31930022, 81911530167, 81870187)
  • Strategic Priority Research Program of the Chinese Academy of Sciences (XDB38040400)
  • National Key R&D Program of China (2022YFA1004800, 2021YFF0703802)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene expression and cancer classification
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Spatial and Panel Data Analysis
Keywords
spatial transcriptomics, multimodal data integration, spatial domain identification, graph convolutional networks, attention mechanism, tumor invasion, colorectal liver metastases, data denoising, single-cell trajectory analysis, 3D feature extraction
MeSH terms
Algorithms, Humans, Liver Neoplasms, Neoplasm Invasiveness, Colorectal Neoplasms, Gene Expression Regulation, Neoplastic, Computational Biology, Gene Expression Profiling, Transcriptome

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.