Spatial Data Integration

Integrates data from fundamentally different spatial platforms—such as Visium (capture-based) and MERFISH (imaging-based)—into a common latent space. This process corrects for technical batch effects while preserving true biological variance, allowing researchers to combine the broad capture areas of array-based methods with the single-cell resolution of in situ techniques.

Co-registers spatial transcriptomics data with complementary modalities from adjacent tissue sections, including spatial proteomics, epigenomics, and H&E histology. Advanced algorithms perform non-rigid image registration to warp coordinate systems, enabling the joint analysis of mRNA expression, chromatin accessibility, and protein abundance within the exact same tissue architecture.

Establishes a unified spatial reference system by mapping all datasets onto a canonical tissue atlas. This involves:

• Landmark-based registration using tissue morphology
• Mutual nearest neighbor (MNN) matching in gene expression space
• Elastic deformation models to account for tissue tearing and folding

Once aligned, queries can be performed across the entire integrated corpus as if it were a single experiment.

Employs specialized algorithms like Harmony, scVI, and LIGER to project multi-sample spatial data into a shared low-dimensional embedding. These methods explicitly model sample-of-origin as a nuisance variable, ensuring that cells cluster by biological identity rather than experimental provenance. The result is a batch-corrected manifold where spatial gradients and gene programs are directly comparable across donors and conditions.

Leverages label transfer algorithms to project annotations from a well-characterized reference dataset onto newly acquired spatial samples. Using techniques like Seurat's anchor-based mapping or Tangram, cell-type identities, gene imputation values, and functional states are mapped onto fresh tissue slices without requiring de novo clustering. This enables rapid, consistent annotation across large cohort studies.

Builds a unified spatial neighborhood graph across integrated datasets by connecting spots or cells based on their aligned physical coordinates. This graph serves as the computational backbone for downstream analyses, including spatial autocorrelation testing, ligand-receptor co-localization across technologies, and spatial trajectory inference that traces differentiation paths through the harmonized tissue architecture.

The computational alignment of multiple tissue images or spatial datasets into a common coordinate system. This process is the foundational prerequisite for integration, enabling direct comparison of gene expression, histology, and protein data from the same or adjacent tissue sections. Without precise registration, cross-modality analysis is impossible.

• Landmark-based: Uses manual or automated detection of corresponding tissue features.
• Intensity-based: Optimizes alignment by maximizing pixel/voxel similarity.
• Deformable: Applies non-linear warping to account for tissue distortion.

A computational method for removing non-biological technical variation between multiple spatial transcriptomic samples or experiments. Unlike standard single-cell batch correction, these algorithms must preserve the true biological spatial heterogeneity and tissue architecture. Failure to correct for batch effects leads to false spatial domains and incorrect ligand-receptor inferences.

• Harmony-Spatial: Adapts Harmony's soft-clustering for spatial covariance.
• SpatialMNN: Extends mutual nearest neighbors to spatial coordinates.
• ComBat-Seq: A negative binomial regression approach for count data.

The computational fusion of spatial transcriptomics with other spatially resolved modalities from the same or adjacent tissue sections. This includes co-registering mRNA data with immunofluorescence protein markers, spatial epigenomics (ATAC-seq), or spatial metabolomics. The goal is to build a unified molecular view of a tissue niche that no single assay can provide.

• Seurat v5: Supports bridge integration and multi-modal reference mapping.
• MultiVI: A probabilistic model for joint analysis of paired multi-omics data.
• SpatialGlue: A graph neural network approach for cross-modality fusion.

A data structure where each spatial location (cell, spot, or bead) is a node, and edges connect neighboring locations based on a distance threshold or k-nearest neighbors. This graph is the computational backbone for spatial-aware algorithms, enabling message passing in graph neural networks and defining the covariance structure for spatial statistics.

• Delaunay triangulation: Creates edges based on Voronoi tessellation.
• Radius graph: Connects all spots within a fixed physical distance.
• k-NN graph: Connects each spot to its k-closest spatial neighbors.

A computational technique for predicting the expression of unmeasured genes or enhancing the resolution of sparse spatial transcriptomics data. By leveraging gene-gene correlations and spatial autocorrelation, these models can computationally increase the effective resolution of a dataset, turning low-resolution spot-based data into a subcellular-like map.

• Tangram: Aligns scRNA-seq data onto spatial coordinates using deep learning.
• SpaGE: Links spatial and scRNA-seq data via a domain adaptation framework.
• Seurat Transfer: Uses canonical correlation analysis to project reference data.

The unsupervised identification of tissue regions with coherent gene expression profiles and histology. These algorithms partition a tissue into anatomically meaningful zones without prior knowledge, often revealing novel micro-anatomical structures. This is a critical step for defining regions of interest before performing differential expression analysis.

• BayesSpace: Uses a spatial hidden Markov model for cluster refinement.
• SpaGCN: A graph convolutional network that integrates histology and expression.
• STAGATE: A graph attention auto-encoder for spatial domain identification.