SpatialJEPA is introduced, a JEPA-inspired teacher–student framework for transferring spatial context from spatial multiomics data to non-spatial multiome data, which masks spatial context by replacing the teacher’s spatial neighborhood graph with a self-only identity graph during student training, making the spatial sample appear dissociated to the student.
Experiments demonstrate that the GraphFusionNN fusion strategy— integrating graph topology, spatial context, and external embeddings—significantly improves classification accuracy, macro-F1, and robustness compared to single-modality models.
LATTICE demonstrated stable optimization behavior, reproducible embeddings across analysis seeds, and complete multimodal integration across all samples, position LATTICE as a practical and empirically grounded framework for multimodal spatial omics integration, while also highlighting the need for stronger supervision and broader external benchmarking.
J. Dwarampudi, V. Kochat, Suresh Satpati et al.· 0 citations
Spatial omic technologies have revolutionized tissue analysis by enabling multimodal molecular coprofiling within their native tissue context. Integrating multislice spatial multiomic data offers unprecedented opportunities to reconstruct three-dimensional (3D) tissue landscapes from multimodal molecular perspectives. However, current spatial omic integration methods remain narrowly focused on either vertical (cross-omic) or horizontal (cross-slice) integration, leaving a critical gap for a unified framework that simultaneously addresses both dimensions. Here we present SpatialMOSI, a unified framework for mosaic integration that concurrently resolves cross-modality and cross-section variations. At its core, SpatialMOSI employs a hierarchical graph contrastive learning (HiGCL) strategy that coordinates three integrative objectives: cross-omic alignment and fusion, cross-slice batch correction, and spatial microenvironment preservation. This approach operates on modality-specific latent representations while maintaining feature fidelity through decoding reconstruction. We demonstrate SpatialMOSI's versatility across multiple biological systems, accurately identifying spatially conserved domains, imputing missing omic layers, revealing B cell dynamics in germinal centers, delineating tumor-immune interactions, and reconstructing embryonic developmental trajectories. SpatialMOSI provides a critical computational foundation for constructing integrative 3D molecular atlases from complex multimodal spatial data sets.
Peimeng Zhen, Han Shu, Bingtao Wang et al.· Genome Research· 0 citations
Spatial transcriptomics enables the study of gene expression within its native tissue context, providing critical insights into cellular organization and microenvironment-driven biological processes. A key challenge in this field is spatial domain identification, which aims to partition tissue into coherent regions by jointly leveraging gene expression and spatial information. Existing approaches are predominantly based on Graph Neural Networks (GNNs), and approach based on Transformers particularly, Bidirectional Encoder Reppresentation Transformer (BERT) model for modelling both local and long-range dependencies remains largely unexplored. In this work, we propose BERT for Spatial Transcriptomics (BertST), a transformer-based framework that reformulates spatial transcriptomics as a graph-to-text representation learning problem. Building upon the BERTwalk paradigm, we construct a task-specific multi-graph representation integrating spatial adjacency, pruned gene-expression similarity, and a fully connected gene-expression graph. This design enables the modelling of both local spatial structure and global molecular relationships. Random walks over these graphs are treated as sequences, allowing a BERT model to learn contextualised node embeddings. To further enhance representation quality, we introduce a hierarchical multi-graph propagation strategy, where embedding refinement is performed sequentially: first on the fully connected graph to capture global structure, followed by the pruned graph to refine molecular relationships, and finally on the spatial graph to enforce local smoothness. This ordering ensures that global information is effectively distributed and progressively constrained by biologically meaningful neighbourhoods. We also improve computational efficiency by leveraging PecanPy, a fast and scalable implementation of node2vec, enabling efficient random walk generation on dense graphs. Experimental results on multiple 10x Visium datasets, including DLPFC and Human Breast Cancer, demonstrate that BertST consistently outperforms or matches GNN-based methods such as ConST, CCST, and SpaceFlow in terms of Adjusted Rand Index (ARI) and Adjusted Mutual Information (AMI). Overall, BertST highlights the potential of transformer-based architectures for spatial omics analysis by effectively capturing both local and long-range spatial-molecular dependencies, offering a promising alternative to traditional graph-based methods.
Multimodal spatial omics integration offers a powerful paradigm to decipher the hierarchical regulatory mechanisms underlying cellular function and tissue architecture. In this study, a novel method of multimodal spatial omics fusion, named mmspao, is proposed to obtain cross-modal interactive features and combine them with the features of each modality to obtain fusion results at the spatial resolution. This method integrates the data of spatial transcriptomics, epigenomics, and proteomics. It further integrates information from each modality using adjacency graph modeling and latent spatial representation. We demonstrate the effectiveness of this method on simulated and real multiomics data. By leveraging adjacency graph modeling and latent spatial representation, mmspao effectively aligns multiomics modalities within shared spatial domains, preserving individual gene expression profiles while generating globally integrated fusion maps that advance the decoding of tissue architecture and cellular regulatory hierarchies.
Spatial transcriptomics measures gene expression together with spatial locations, but its data are noisy and sparse, and existing graph-based methods are complex and hard to scale.
We present AINR, an end-to-end deep learning framework that models spatial transcriptomics data as a geometrically constrained continuous biological field. AINR combines implicit neural representations with a spatially-aware attention mechanism and a total variation regularization term, using a periodic sine activation function to map spatial coordinates directly to gene expression while preserving spatial smoothness without explicit adjacency matrices. Across six diverse datasets, AINR consistently outperforms existing methods in spatial domain identification and remains robust even under extreme data sparsity.
The code for AINR is available at https://github.com/XGD1122/AINR
Yusen Zhang, Guodong Xiao, Ponian Li et al.· Bioinformatics Advances· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
