stMMR: accurate and robust spatial domain identification from spatially resolved transcriptomics with multi-modal feature representation

This article has been Reviewed by the following groups

Read the full article See related articles

Abstract

Deciphering spatial domains using spatially resolved transcriptomics (SRT) is of great value for the characterizing and understanding of tissue architecture. However, the inherent heterogeneity and varying spatial resolutions present challenges in the joint analysis of multi-modal SRT data. We introduce a multi-modal geometric deep learning method, named stMMR, to effectively integrate gene expression, spatial location and histological information for accurate identifying spatial domains from SRT data. stMMR uses graph convolutional networks (GCN) and self-attention module for deep embedding of features within unimodal and incorporates similarity contrastive learning for integrating features across modalities. Comprehensive benchmark analysis on various types of spatial data shows superior performance of stMMR in multiple analyses, including spatial domain identification, pseudo-spatiotemporal analysis, and domain-specific gene discovery. In chicken heart development, stMMR reconstruct the spatiotemporal lineage structures indicating accurate developmental sequence. In breast cancer and lung cancer, stMMR clearly delineated the tumor microenvironment and identified marker genes associated with diagnosis and prognosis. Overall, stMMR is capable of effectively utilizing the multi-modal information of various SRT data to explore and characterize tissue architectures of homeostasis, development and tumor.

Article activity feed

  1. performance of stMMR in multiple analyses, including spatial domain identification, pseudo-spatiotemporal analysis, and domain-specific gene discovery. In chicken heart development, stMMR reconstruct the spatiotemporal lineage structures indicating accurate developmental sequence. In breast cancer and lung cancer, stMMR clearly delineated the tumor microenvironment and identified marker genes associated with diagnosis and prognosis. Overall, stMMR is capable of effectively utilizing the multi-modal information of various SRT data to explore and characterize tissue architectures of homeostasis, development and tumor.

    Reviewer 2: Hongzhi Wen Reviewer Comments: The paper introduces stMMR, a multi-modal graph learning method designed to integrate gene expression, spatial location, and histological information for accurate spatial domain identification from spatially resolved transcriptomics (SRT) data. The method employs graph convolutional networks (GCN) and self-attention modules, along with cross-modal contrastive learning, to enhance feature integration and representation.Strengths:1. Using GCN to capture local spatial dependency is natural and effective. Introducing attention mechanism for capturing global relations intuitively make senses, however, need more justification. Contrastive learning for cross-modal feature fusion is also a natural choice in multimodal learning. Overall, the methodology is novel and solid.2. Extensive benchmark analysis across various types of spatial data and tissues demonstrates superior performance of the method in spatial domain identification, pseudo-spatiotemporal analysis, and domain-specific gene discovery. The empirical evidence is very convincing.3. The method's application to chicken heart development, breast cancer, and lung cancer showcases its potential in reconstructing spatiotemporal lineage structures and delineating tumor microenvironments, highlighting its value in clinical research.Weaknesses:1. In Figure 4, SpaceFlow is the only baseline for the case study. However, the performance of SpaceFlow is not topranked in other experiments. There should be a justification for why SpaceFlow is highlighted here.2. The contribution of the global attention mechanism to the whole framework is not very clear. The authors may provide more intuition and empirical justification (e.g., ablation study) if they would like to highlight this design.3. By introducing the hyperparameters $\alpha$, $\beta$ and $\gamma$ in Eq. (11), the method has a significantly larger search space than other methods. It is important to note how these hyperparameters are chosen in practice, more importantly, whether the test performance is referred when adjusting these hyperparameters. This might result in an unfair evaluation.

  2. AbstractDeciphering spatial domains using spatially resolved transcriptomics (SRT) is of great value for the characterizing and understanding of tissue architecture. However, the inherent heterogeneity and varying spatial resolutions present challenges in the joint analysis of multi-modal SRT data. We introduce a multi-modal geometric deep learning method, named stMMR, to effectively integrate gene expression, spatial location and histological information for accurate identifying spatial domains from SRT data. stMMR uses graph convolutional networks (GCN) and self-attention module for deep embedding of features within unimodal and incorporates similarity contrastive learning for integrating features across modalities. Comprehensive benchmark analysis on various types of spatial data shows superior

    Reviewer 1: Shihua Zhang Reviewer Comments: In this paper, the authors developed a multi-modal deep learning method for identifying spatial domains from ST data by integrating gene expression, spatial location and histological information. This method adopts the graphconvolutional networks and self-attention module for deep embedding of features within unimodal and incorporates similarity contrastive learning for integrating features across modalities. They did several typical analysis to valid this this method. Generally, the wiring of this paper is OK. More specific comments:1. Spatial domain has been overwhelmingly studied recently. The authors need to pay more attention to why it is needed to introduce a new method. The novelty of the current method should be carefully clarified. For example, how the histological information help to improve the performance? Does the "geometric" deep learning really help?2. This method has been applied to some stereotypical data. The authors should applied it to some recently generated data by some new ST techniques.3. Figure 3 stMMR enhances spatial gene expression profiles. It is hard to see how the method enhance the spatial gene expression (e.g., LPL).4. With the accumulation of multi-slice spatial transcriptome data, the integration and alignment of spatial transcriptome data will be essential. Can this method be extended for this situation like STAGATE (Nat Comput Sci.2023 Oct; 3(10):894-906)? This will be valuable for ST analysis.5. The scalability of this method should be carefully explored.6. The authors should provide a detailed tutorial for users.