Spatial Clustering and Cell-Type Mapping

  • Spatial clustering and cell-type mapping are important computational approaches in spatial transcriptomics and spatial omics. They help researchers identify groups of cells or tissue regions with similar molecular characteristics while preserving information about their physical locations. Spatial clustering identifies spatially organized groups based on gene expression, molecular profiles, spatial proximity, or combinations of these features. Cell-type mapping assigns biological identities to individual cells, spatial spots, or tissue regions by comparing their molecular characteristics with known cell-type markers or reference datasets. Together, these approaches help reveal tissue organization, cellular diversity, developmental patterns, disease-associated changes, and interactions between neighbouring cell populations.
  • Spatial clustering differs from conventional clustering because it considers the spatial relationships between observations in addition to their molecular similarities. In conventional single-cell RNA sequencing, cells are commonly grouped according to similarities in their gene-expression profiles, without considering their original tissue locations. In spatial transcriptomics, each cell or spot has associated spatial coordinates, allowing researchers to investigate whether molecularly similar observations are also physically close to one another. This additional information can reveal organized tissue compartments, cellular neighbourhoods, spatial gradients, and localized biological processes.
  • Spatial clustering can be performed using gene-expression profiles, spatial coordinates, or integrated molecular and spatial information. Expression-based clustering groups cells or spots according to similarities in their transcript abundance. Spatial-coordinate clustering identifies groups of observations that are physically close to each other, regardless of whether their expression profiles are identical. Joint spatial clustering combines molecular similarity and spatial proximity to identify groups that share both biological characteristics and tissue localization. The choice of approach depends on the research objective because expression similarity and spatial proximity do not always represent the same biological relationship.
  • A common starting point is to construct a spatial neighbourhood graph. In this graph, each cell or spot is represented as a node, and edges connect observations that are considered spatial neighbours. Neighbours may be defined using physical distance, a fixed number of nearest observations, or platform-specific tissue geometry. For spot-based technologies, adjacency may be based on the arrangement of capture spots. For single-cell-resolution datasets, neighbouring cells can be identified using their spatial coordinates or segmented cell boundaries. The resulting graph provides a framework for analysing spatial relationships and applying graph-based clustering algorithms.
  • Several clustering algorithms are used in spatial transcriptomics. The Leiden and Louvain algorithms identify communities within neighbourhood graphs and are widely used in single-cell and spatial expression analysis. Hierarchical clustering groups observations according to their molecular similarity at different levels of resolution. K-means clustering divides observations into a predefined number of groups based on their feature values. Density-based methods, such as DBSCAN, identify dense groups of observations and can distinguish isolated observations from spatially concentrated populations. Graph-based methods are particularly useful when spatial and molecular relationships need to be analysed together.
  • The spatial resolution of a dataset affects clustering results. In spot-based spatial transcriptomics, each spot may contain transcripts from several cells, producing mixed expression profiles. Consequently, clusters may represent tissue regions or mixtures of cell populations rather than individual cell types. In single-cell-resolution technologies such as Xenium, CosMx, MERFISH, and seqFISH, clustering can be performed at the individual-cell level when segmentation and transcript assignment are sufficiently accurate. However, high spatial resolution does not automatically guarantee correct cell identification, because incomplete transcript detection and segmentation errors can still affect molecular profiles.
  • Spatial clustering often follows quality control, normalization, feature selection, and dimensionality reduction. Highly variable genes or biologically informative genes may be selected to represent differences between cells or spots. Principal component analysis (PCA) can reduce the dimensionality of gene-expression data before construction of neighbourhood graphs. Spatial coordinates can then be incorporated into graph construction or clustering models. Depending on the method, molecular features and spatial information may be weighted differently. This is important because excessive emphasis on spatial proximity may group neighbouring but biologically distinct cells, while ignoring spatial information may fail to identify meaningful tissue organization.
  • Several computational tools support spatial clustering. Seurat provides functions for processing spatial transcriptomics datasets, constructing neighbourhood graphs, clustering observations, and visualizing spatial expression patterns. Scanpy supports expression-based analysis and graph clustering, while Squidpy extends the Scanpy ecosystem with spatial neighbourhood graphs, spatial statistics, and tissue-level analysis. Giotto provides an integrated framework for spatial transcriptomics, including spatial network construction, clustering, and visualization. SpatialExperiment offers a structured Bioconductor framework for managing spatial molecular data. Other approaches, including BayesSpace, SpaGCN, stLearn, and BANKSY, provide specialized methods for identifying spatial domains or integrating gene-expression and spatial information.
  • Spatial domain identification is closely related to spatial clustering. A spatial domain is a tissue region characterized by a relatively consistent molecular profile and spatial organization. Domains may correspond to anatomical structures, functional tissue compartments, tumour regions, or areas with distinct cellular compositions. For example, spatial analysis of a tumour may identify malignant regions, stromal compartments, immune-cell-rich areas, and regions with low immune-cell infiltration. In brain tissue, spatial domains may correspond to anatomical regions containing distinct neuronal and glial populations. Identifying these domains helps researchers understand how molecular activity varies across tissue architecture.
  • Cell-type mapping is the process of assigning biological identities to cells or spatial locations based on their molecular characteristics. It is an essential step because spatial transcriptomics initially produces gene-expression measurements rather than definitive biological cell labels. Cell-type identification commonly uses marker genes, reference single-cell datasets, computational classification, and biological knowledge. Marker genes are genes that are enriched in particular cell populations and help distinguish one cell type from another. However, marker expression is not always exclusive to a single cell type, so several markers are generally considered together rather than relying on one gene.
  • Common marker genes provide examples of how cell-type mapping is performed. CD3D and CD3E are associated with T lymphocytes, while MS4A1 is commonly used as a B-cell marker. LST1 and LYZ are frequently associated with myeloid populations. EPCAM is a widely used epithelial-cell marker, PECAM1 is associated with endothelial cells, and COL1A1 and DCN are commonly expressed in fibroblast populations. In muscle tissues, MYOD1 and MYOG are associated with muscle differentiation and myogenic activity. These markers can support cell-type annotation, but their expression must be interpreted in the context of tissue, species, developmental stage, and cellular state.
  • Marker-based annotation generally involves identifying genes that are enriched in a cluster and comparing them with established cell-type markers. Differential expression tools such as Seurat FindMarkers and FindAllMarkers, Scanpy rank_genes_groups, MAST, edgeR, and DESeq2 can help identify cluster-associated genes, depending on the data structure and statistical design. Gene-expression patterns can then be compared with curated marker resources such as CellMarker, PanglaoDB, the Human Protein Atlas, and the Cell Ontology. Functional enrichment using Gene Ontology, KEGG, and Reactome can provide additional evidence about the biological roles of identified cell populations.
  • Reference-based cell-type mapping uses an independently annotated single-cell RNA sequencing dataset as a reference. The expression profiles of spatial cells or spots are compared with known reference cell types to assign labels or estimate cellular composition. Important tools include Seurat reference mapping, SingleR, CellTypist, scmap, scArches, Azimuth, and Garnett. These methods differ in their classification strategies, reference requirements, and ability to transfer annotations between datasets. Reference-based mapping is particularly useful when spatial data have limited gene coverage or when cell types are difficult to distinguish using a small number of marker genes.
  • An important challenge in spot-based spatial transcriptomics is that each spot may contain multiple cell types. In such cases, assigning one cell-type label to an entire spot may oversimplify its biological composition. Spatial deconvolution methods estimate the proportions or contributions of different cell types within each spot by using reference single-cell expression profiles. RCTD, cell2location, SPOTlight, stereoscope, DestVI, and Tangram are important computational approaches for this purpose. These methods help reconstruct the likely cellular composition of tissue regions, although their estimates depend on reference quality, gene coverage, cell-type diversity, and the assumptions of the underlying models. Estimated cell proportions should not be interpreted as direct observations of individually identified cells.
  • Cell-type mapping can also be performed through spatial reference mapping, in which annotated single-cell profiles are matched to locations in a tissue. This approach helps transfer cell identities from single-cell RNA sequencing to spatial transcriptomics datasets. It is particularly valuable when spatial platforms measure fewer genes than whole-transcriptome single-cell sequencing. However, differences in tissue preparation, sequencing depth, gene detection, biological conditions, and species can affect mapping accuracy. Cross-species mapping requires particular care because genes may have different expression patterns, and orthologous genes may not perform identical biological functions in different organisms.
  • After cell-type mapping, researchers can examine the spatial distribution of specific populations. Spatial plots can show where immune cells, epithelial cells, fibroblasts, endothelial cells, neurons, or other populations are located within a tissue. Spatial neighbourhood analysis investigates which cell types occur near one another and whether particular combinations of populations are enriched in specific regions. These analyses can reveal tissue organization, immune infiltration, cellular barriers, and potential interactions between neighbouring cell populations. Tools such as Squidpy, Giotto, Seurat, and SpatialExperiment support different aspects of spatial visualization and neighbourhood analysis.
  • Spatial clustering and cell-type mapping can also be integrated with spatially variable gene analysis. Spatially variable genes are genes whose expression changes systematically across tissue locations rather than being distributed randomly. Methods such as SpatialDE, SPARK, and SPARK-X can identify genes with spatially structured expression patterns. These genes may help characterize anatomical regions, developmental gradients, tumour microenvironments, and localized biological activities. Combining spatial clustering with spatially variable gene analysis can provide a more complete description of molecular tissue organization.
  • An important extension is the analysis of cell–cell communication. Once cell types and their spatial relationships have been established, researchers can investigate potential communication between nearby populations through ligand–receptor interactions. CellChat, CellPhoneDB, NicheNet, and LIANA can help identify candidate communication pathways using gene-expression information. Spatial proximity can provide additional biological context because neighbouring cells are more likely to participate in some forms of local signalling. However, transcript-based ligand–receptor predictions do not prove that a signalling interaction occurs. Protein abundance, receptor activation, secretion, and experimental validation should be considered.
  • Spatial clustering and cell-type mapping have applications in cancer research, neuroscience, developmental biology, immunology, infectious disease studies, and regenerative medicine. In cancer, they can distinguish tumour cells, immune populations, fibroblasts, endothelial cells, and other components of the tumour microenvironment. In neuroscience, they help identify neuronal and glial populations and map their organization across brain regions. In developmental biology, they reveal how cell populations are distributed during tissue formation and differentiation. In immunology, they can characterize immune-cell infiltration and the organization of local immune environments.
  • In animal genomics and breeding research, spatial clustering and cell-type mapping can be used to investigate cellular organization in skeletal muscle, mammary glands, reproductive organs, liver, and immune tissues. They may help identify cell populations associated with muscle growth, muscle-fibre development, meat quality, milk production, fertility, disease resistance, and adaptation to environmental conditions. Comparisons between breeds, developmental stages, physiological conditions, and disease states may reveal differences in cellular composition and spatial gene-expression patterns that contribute to economically important traits.
  • Several limitations must be considered when interpreting spatial clustering and cell-type mapping results. Cluster boundaries may depend on the selected genes, normalization method, spatial graph, algorithm, and resolution parameters. Different algorithms can produce different cluster structures from the same dataset. Marker genes may be shared between cell types or vary across developmental stages and physiological conditions. Reference datasets may not contain all cell populations present in the tissue, and incomplete reference representation can lead to incorrect annotations. Spot-based datasets add uncertainty because multiple cells contribute to a single expression profile. Technical artefacts, batch effects, tissue damage, segmentation errors, and low transcript detection can further affect the results.
  • Reliable interpretation therefore requires combining computational predictions with biological knowledge and independent validation. Cell-type assignments can be examined using multiple marker genes, reference mapping, histological images, immunofluorescence, immunohistochemistry, RNA in situ hybridization, and other experimental approaches. Biological replicates are important when comparing conditions, and statistical analysis should account for the experimental unit rather than treating every cell or spot as an independent biological replicate. Researchers should also report the clustering method, spatial-neighbourhood definition, resolution settings, reference dataset, annotation criteria, and filtering procedures.
  • Overall, spatial clustering and cell-type mapping are essential for understanding how molecularly distinct cell populations are organized within tissues. Spatial clustering identifies tissue domains and groups of related observations, while cell-type mapping assigns biological identities to cells or estimates cellular composition within spatial locations. Their integration with spatial deconvolution, spatially variable gene analysis, cell–cell communication, and multi-omics data provides a powerful framework for investigating tissue architecture and cellular function. These approaches are increasingly important in biomedical research, veterinary science, and animal genomics because they connect molecular information with the physical organization of biological tissues.
Author:

Leave a Reply

Your email address will not be published. Required fields are marked *