Spatial Multi-Omics Integration

The foundational step of aligning tissue sections assayed with different technologies into a common coordinate framework. This process corrects for physical deformations, rotation, and scaling differences between serial sections. Landmark-based registration uses histology images to find corresponding tissue features, while intensity-based methods optimize pixel-level similarity. Successful registration is critical for enabling direct, pixel-to-pixel correlation of transcriptomic and proteomic signals.

A family of dimensionality reduction techniques designed to decompose integrated spatial data into a shared latent space. Methods like Multi-Omics Factor Analysis (MOFA) and Spatial Multi-Omics Variational Autoencoders learn factors that explain variance across all data modalities simultaneously. This reveals coordinated multi-omic signatures of specific tissue niches, such as a factor capturing both gene expression and protein abundance changes at the tumor-immune boundary.

Deep learning architectures that model tissue as a graph, where nodes are spatial locations and edges represent proximity. These models learn to fuse features from different omics layers by passing messages between neighboring nodes. A spatial graph convolutional network can predict the expression of a protein at a spot where only mRNA was measured, effectively imputing missing modalities by learning the relationship between local transcriptomic and proteomic states.

Computational strategies that use one measured modality to predict another unmeasured modality on the same or an adjacent section. For example, using spatial transcriptomics data to predict chromatin accessibility states or protein abundance at each spatial location. This is often achieved through paired training datasets and models like conditional generative adversarial networks (cGANs) , enabling the construction of a virtual multi-omic atlas from a single experimental input.

Statistical frameworks that move beyond simple gene-gene correlation to quantify the spatial relationship between different molecular species. This includes cross-correlation functions that measure how the spatial pattern of a protein relates to the pattern of a transcript. A key application is validating ligand-receptor interactions by requiring that the mRNA of a ligand, the protein of its receptor, and their spatial proximity all align within a defined tissue microdomain.

Unsupervised clustering algorithms that partition tissue into functional regions using all available data layers. Unlike single-omic domain detection, this approach identifies regions defined by a consensus of transcriptomic, proteomic, and epigenomic states. A joint spatial hidden Markov model can infer domains where gene expression, protein activity, and chromatin state all shift coherently, revealing tissue architecture invisible to any single assay.

The computational alignment of multiple tissue images or spatial datasets into a common coordinate system. This is the foundational preprocessing step for multi-omics integration, enabling the direct comparison of transcriptomic signals from one section with proteomic signals from an adjacent section. Techniques range from landmark-based affine transformations to non-linear elastic deformations that warp images to account for tissue tearing or stretching during sectioning.

A computational process that estimates the proportions of different cell types within a spatial transcriptomics spot by separating the mixed gene expression signal. In multi-omics integration, deconvolution results are often cross-validated against spatially resolved proteomics data (e.g., CODEX or MIBI) to confirm the presence and abundance of inferred cell populations, providing a more robust cellular map.

The unsupervised identification of tissue regions with coherent gene expression profiles and histology. When integrating multi-omics data, domains defined by transcriptomic clustering can be overlaid with chromatin accessibility (from spatial ATAC-seq) or protein abundance maps to characterize the regulatory and functional state of each anatomical niche. Common methods include graph-based clustering and spatial hidden Markov models.

A deep learning architecture that operates on graph representations of spatial data, where nodes represent cells or spots and edges represent spatial proximity. These models are uniquely suited for multi-omics integration because they can learn a joint latent space from disparate input modalities (e.g., gene expression and protein intensity) while explicitly accounting for the tissue's spatial neighborhood structure.

A computational analysis that identifies spatially proximal cell-type pairs where a ligand gene in one cell type and its cognate receptor gene in another are co-expressed. Multi-omics integration strengthens this analysis by confirming the presence of the ligand and receptor proteins via spatial proteomics, moving beyond transcript-level inference to validate functional cell-cell communication axes within the tissue microenvironment.

A computational method for removing technical variation between multiple spatial transcriptomic samples or experiments while preserving true biological spatial heterogeneity. This is critical for multi-omics studies where data is generated across different platforms (e.g., Visium for transcriptomics and MIBI for proteomics) and must be harmonized to remove platform-specific artifacts before joint analysis.