Determinants of Interest-Holder Perspectives Toward Post-Mortem Tissue Donation in Oncology: A Systematic Review.

Spatial heterogeneity is a fundamental property of complex tissues, manifested as regional differences in cellular composition, transcriptional programmes, and tissue architecture. Such heterogeneity underlies functional specialization within tissues and plays a central role in diverse biological processes, including development, tissue homeostasis, and disease progression. Therefore, systematically characterizing spatial heterogeneity is essential for understanding how tissue organization relates to biological function. Recent spatial transcriptomics (ST) technologies (, ), such as 10× Visium (, ), Stereo-seq (), and multiplexed error-robust fluorescence in situ hybridization (MERFISH) (), have created unprecedented opportunities to address this challenge. An important step toward characterizing spatial heterogeneity in ST data is spatial domain identification, which aims to partition tissue into regions with coherent molecular and structural characteristics. Current spatial domain identification methods can be broadly classified into two categories, namely non-spatial clustering methods and spatial clustering methods. Non-spatial clustering methods, such as K-means, Seurat (), and Louvain (), rely primarily on gene expression profiles while neglecting spatial context. Consequently, these approaches often struggle to accurately reconstruct continuous tissue structures. To overcome this limitation, spatial clustering methods like STAGATE (), GraphST (), and SpatialGEO () leverage the spatial coordinates to aggregate neighbourhood information using graph neural networks (GNNs) or statistical models. By exploiting spatial proximity, these approaches enforce spatial coherence, facilitating a more precise delineation of tissue microenvironments. In addition, multimodal ST has made it possible to decipher spatial heterogeneity by integrating diverse data modalities within the original tissue sections. Some ST technologies provide matched histological images alongside gene expression profiles and spatial coordinates, most commonly haematoxylin and eosin (H&E) staining. These images provide important morphological information, including cellular morphology, tissue density, and structural boundaries, thereby offering complementary information about tissue architecture and microenvironmental organization. Consequently, the effective integration of multimodal ST data is crucial for accurately deciphering spatial heterogeneity in tissues. Several methods, including SpaGCN (), DeepST (), stLearn (), and STAIG (), have attempted to incorporate multiple modalities for spatial heterogeneity analysis. However, in many existing approaches, image features are often treated primarily as auxiliary information to refine spatial graphs, guide denoising, or impose additional regularization. Image features are not deeply integrated with gene features and spatial information to form a joint representation. To address these challenges, we present SpatialModal, a multimodal graph learning framework that learns robust joint representations by combining a hierarchical representation strategy with a dual-level contrastive learning mechanism. We extensively validate the performance of SpatialModal across diverse ST datasets spanning various tissue types in humans and mice. The results demonstrate that SpatialModal effectively reveals intricate brain structures, tumour microenvironment heterogeneity, Alzheimer’s disease patterns, and spatiotemporal developmental trajectories within the embryonic heart, highlighting its capability to decipher the spatial intricacies of biological tissues. Furthermore, although SpatialModal is designed for multimodal ST data, its architecture can naturally adapt to settings where histological images are unavailable.