Histopathology-centered Computational Evolution of Spatial Omics: Integration, Mapping, and Foundation Models

TL;DR

The paper tackles the challenge of integrating morphological histopathology with spatial omics by proposing a histopathology-centered framework that unifies integration, mapping, and foundation-model paradigms. It analyzes how H&E morphology evolves from a contextual backdrop to a predictive anchor and ultimately to a representation backbone, highlighting data limitations and resolution demands that drive methodological shifts. Key contributions include a structured taxonomy of methods, up-to-date coverage of spatial foundation models across ST, SP, and beyond, and a critical discussion of common failures with actionable directions. The work provides a practical roadmap for researchers and clinicians to harness H&E-centered multimodal learning to achieve cellular-level insights and clinically relevant endpoints in spatial omics.

Abstract

Spatial omics (SO) technologies enable spatially resolved molecular profiling, while hematoxylin and eosin (H&E) imaging remains the gold standard for morphological assessment in clinical pathology. Recent computational advances increasingly place H&E images at the center of SO analysis, bridging morphology with transcriptomic, proteomic, and other spatial molecular modalities, and pushing resolution toward the single-cell level. In this survey, we systematically review the computational evolution of SO from a histopathology-centered perspective and organize existing methods into three paradigms: integration, which jointly models paired multimodal data; mapping, which infers molecular profiles from H&E images; and foundation models, which learn generalizable representations from large-scale spatial datasets. We analyze how the role of H&E images evolves across these paradigms from spatial context to predictive anchor and ultimately to representation backbone in response to practical constraints such as limited paired data and increasing resolution demands. We further summarize actionable modeling directions enabled by current architectures and delineate persistent gaps driven by data, biology, and technology that are unlikely to be resolved by model design alone. Together, this survey provides a histopathology-centered roadmap for developing and applying computational frameworks in SO.

Figures (4)

• Figure 1: Computational landscape of spatial omics and H&E image analysis. a. Illustration of general ST and SP techniques. b. Commonly used public SO datasets. Details are in the Additional file 1: Table S1. c. Three categories of current methods: Integration methods align and fusion multiple data modalities; Mapping methods focus on H&E-to-X prediction or imputation; Foundation models are pretrained on large heterogeneous datasets and fine-tuned for spatial omics tasks. Additional file 1: Table S2 presents details of all reviewed methods. d. Example architectures used in integration, mapping, and foundation models. e. Representative downstream tasks. Abbreviations: SO, spatial omics; ST, spatial transcriptomics; SP, spatial proteomics; sc, single-cell; CNN, convolutional neural network; MLP, multilayer perceptron; GNN, graph neural network; GCN, graph convolutional network; ViT, vision transformer; CLIP, contrastive language-image pretraining; DE, differentially expressed.
• Figure 2: A roadmap of computational methods of spatial omics and H&E image analysis: Integration, Mapping, and Foundation Models. Solid boxes denote peer-reviewed publications; dashed boxes indicate preprints. Red boxes denote methods with explicit single-cell-level support (These methods are annotated as 'Yes' in the 'Single-cell level' column in the Additional file 1: Table S2). Some methods that do not explicitly incorporate H&E images (e.g., SpaTrio yang2023revealing) are presented for completeness. Abbreviations: SO, spatial omics; ST, spatial transcriptomics; SP, spatial proteomics; SM, spatial metabolomics; scOmics, single-cell omics; scRNA, single-cell RNA; IHC, immunohistochemistry.
• Figure 3: Integration, Mapping, and Foundation models. Abbreviations: CNN, convolutional neural network; GNN, graph neural network; MLP, multilayer perceptron; ViT, vision transformer; ST, spatial transcriptomics; CLIP, contrastive language-image pretraining.
• Figure 4: Downstream tasks of computational methods. We categorize downstream tasks into five major classes, encompassing spatial architecture identification, functional spatial pattern recognition, molecular profile imputation, cell–cell network inference, and biomedical endpoint prediction.