Spatial Transcriptomics

Unlike dissociation-based single-cell RNA-seq, spatial methods retain the extracellular matrix and cellular neighborhoods. This allows direct correlation of gene expression patterns with histological features like tumor margins, necrotic cores, or lymphoid follicles. The spatial context reveals ligand-receptor interactions that occur only within specific tissue microenvironments.

Technologies like 10x Visium use spatially barcoded oligonucleotide arrays to capture polyadenylated mRNA across entire tissue sections. This hypothesis-free approach maps the complete transcriptome without prior gene selection. The resulting data enables discovery of novel spatial gene expression signatures and unexpected cell-state transitions that targeted panels might miss.

Each spatial capture spot often contains multiple cells. Deconvolution algorithms (e.g., RCTD, SPOTlight, cell2location) integrate scRNA-seq reference data to estimate cell-type proportions per spot. This transforms mixed pixel signals into interpretable cell-type maps, enabling analysis of immune infiltration gradients and tumor-immune interfaces without single-cell resolution hardware.

Statistical frameworks like SpatialDE and SPARK-X identify genes whose expression depends on spatial location, not just cell identity. These spatially variable genes often encode morphogens, guidance cues, or extracellular matrix remodeling factors. Detecting them reveals the molecular gradients that orchestrate tissue patterning during development and disease progression.

Advanced platforms now co-register transcriptomics with protein detection (e.g., CITE-seq spatial, GeoMx DSP) or histological imaging (H&E, immunofluorescence). This multi-modal fusion anchors molecular data to morphological landmarks. Deep learning models like GNNs and vision transformers jointly embed image patches and gene expression vectors, learning features invisible to either modality alone.

Tools like CellChat, NicheNet, and stLearn infer cell-cell communication by analyzing co-expression of ligand and receptor gene pairs within spatial neighborhoods. Unlike dissociated single-cell methods, spatial context constrains which interactions are physically possible. This identifies signaling hubs and juxtacrine interactions that drive tissue homeostasis or tumor immunosuppression.

Single-cell RNA sequencing profiles the transcriptome of individual cells, resolving heterogeneity masked in bulk sequencing. While scRNA-seq dissociates tissue and loses spatial context, it provides the cell-type resolution that spatial transcriptomics maps back onto tissue architecture. Integration of scRNA-seq reference data enables cell-type deconvolution of spatial spots.

Computational alignment of multiple datasets to remove batch effects while preserving true biological variation. In spatial transcriptomics, integration methods like Harmony, scVI, and Seurat's CCA align spatial data with scRNA-seq references, enabling robust cell-type annotation and cross-condition comparisons across tissue sections.

Computational inference of cell-cell communication by mapping ligand expression in one cell type to cognate receptor expression in neighboring cells. Spatial transcriptomics adds physical proximity constraints to these predictions, distinguishing paracrine signaling from endocrine effects. Tools like CellChat and NicheNet incorporate spatial coordinates to validate predicted interactions.

Assignment of biological identity labels to cell clusters or spatial spots using curated marker gene databases and reference mapping. In spatial contexts, annotation must account for mixed cell types within capture spots. Algorithms like RCTD and cell2location perform spatially-aware deconvolution to estimate cell-type proportions at each coordinate.

Fusion of disparate data types into unified representations. Spatial transcriptomics increasingly combines with immunofluorescence, H&E histology, and spatial proteomics. Methods like SpatialGlue and MultiVI jointly model gene expression, protein abundance, and tissue morphology to resolve tissue architecture with multi-omic depth.

Mathematical transformation of high-dimensional spatial data into lower-dimensional space using PCA, t-SNE, or UMAP. Spatial transcriptomics adds the challenge of preserving spatial autocorrelation—the tendency for nearby cells to share transcriptional states. Methods like SpatialPCA explicitly model spatial covariance during embedding.