Spatial Transcriptomics Data Analysis Service
CD Genomics provides a platform-agnostic Spatial Transcriptomics Data Analysis Service that transforms raw spatial expression data into interpreted biological results. Whether your data comes from 10x Visium, Visium HD, Stereo-seq, Xenium, CosMx SMI, or other spatial platforms, our bioinformatics team handles the entire analysis pipeline — from spot/bin/cell-level preprocessing through spatial clustering, domain identification, spatially variable gene detection, and publication-quality visualization. Unlike automated cloud platforms that require you to make analytical decisions, our team designs each analysis step to fit the biological question, sample structure, and data characteristics of your study.
- Platform-agnostic: Visium, Visium HD, Stereo-seq, Xenium, CosMx, and other spatial data formats
- Full analysis pipeline: preprocessing, spatial clustering, domain identification, differential expression, visualization
- Expert-driven, not automated: experienced spatial bioinformaticians design and execute your analysis
- Reproducible output: fully documented code, methods write-up, and publication-ready figures
Data Preprocessing and Quality Control
Spatial transcriptomics data arrives in different formats depending on the platform — spot-level expression matrices (Visium), bin-level data (Stereo-seq, Visium HD), or cell-level feature tables (Xenium, CosMx). Our preprocessing pipeline adapts to each format, starting with rigorous QC before any downstream analysis.
Spot-level and bin-level QC (Visium, Visium HD, Stereo-seq). Raw expression matrices are filtered for spot/bin quality using multiple metrics: detected gene counts per spot, total molecular tag counts, and mitochondrial gene fraction where applicable. Rather than applying fixed universal thresholds, QC cutoffs are data-driven and visually confirmed against tissue histology images — ensuring that filtering does not inadvertently remove biologically meaningful regions with naturally lower transcriptional activity. Tissue regions with folding, tearing, or detachment artifacts are identified from the matched histology image and masked from analysis.
Cell-level QC (Xenium, CosMx SMI). Per-cell metrics — transcript count, gene count, and negative control probe signal — are used to filter low-quality cells. Segmentation quality is assessed spatially: cells at tissue boundaries or in densely packed regions are flagged for quality review. Negative control probes are evaluated per cell to confirm signal specificity.
Normalization and batch correction. Expression values are normalized using SCTransform (spot/bin-level data) or library-size normalization (cell-level data). For multi-sample studies, batch effects are assessed through sample-level clustering and corrected using Harmony or reciprocal PCA integration. Normalization quality is validated by inspecting known marker genes across samples and conditions.
Spatial Clustering and Cell Type Mapping
Spatial clustering identifies transcriptionally similar groups of spots, bins, or cells while incorporating their spatial coordinates — producing clusters that are both molecularly coherent and spatially contiguous.
Spatially aware clustering. Clustering is performed using graph-based methods (Louvain, Leiden) with spatial neighborhood graphs that weight both transcriptomic similarity and physical proximity. For spot/bin-level data, tools such as BayesSpace, SpaGCN, or GraphST are selectively applied depending on data resolution and tissue complexity. For cell-level data, spatial neighborhood analysis is integrated alongside transcriptome-based clustering. Results are visualized as spatial cluster maps overlaid on histology images and as UMAP/t-SNE projections color-coded by cluster and spatial domain.
Cell type annotation. For spot-level data, cell type composition per spot is inferred using cell-type deconvolution (RCTD, SPOTlight, or Cell2location), leveraging matched or reference single-cell RNA-seq datasets. For single-cell-resolution platforms (Xenium, CosMx), cell types are assigned directly based on panel gene expression or through integration with reference atlases. Manual review by our bioinformaticians refines automated calls using canonical marker genes and spatial coherence checks.
Spatial domain identification. Beyond clustering, we identify histologically and transcriptionally coherent tissue domains — tumor core, invasive margin, stroma, lymphoid aggregates, and other anatomically defined regions — using spatially informed segmentation. Domains are annotated by integrating gene signatures, cell-type composition, and histological features.
Spatial Differential Expression Analysis
Spatial differential expression goes beyond comparing predefined groups — it identifies genes whose expression varies across spatial locations, tissue compartments, or conditions, while accounting for spatial autocorrelation.
Spatially variable gene (SVG) detection. Genes whose expression changes systematically across tissue space are identified using spatial statistics (SPARK-X, SpatialDE, or nnSVG). Detected SVGs are ranked by spatial variation significance and visualized as spatial feature plots, allowing researchers to see where in the tissue each gene's expression is highest or lowest.
Compartment-specific differential expression. For studies with defined tissue compartments (e.g., tumor core vs. invasive margin, or immune-infiltrated vs. immune-excluded stroma), differential expression analysis is performed per compartment using pseudobulk aggregation with DESeq2 or limma-based models. This provides statistically grounded lists of genes enriched in each spatial compartment while accounting for within-compartment biological replicates.
Condition-specific spatial comparison. When comparing conditions (treated vs. untreated, disease vs. normal), spatial gene expression patterns are compared across conditions for each tissue compartment. This reveals not just which genes change, but whether the change is spatially restricted — e.g., a gene upregulated in the tumor margin but not the core.
Tissue Region and Structure-Associated Analysis
Spatial data links molecular profiles to tissue architecture. We extract biologically interpretable spatial patterns that connect gene expression to histologically visible structures.
Histology-guided region annotation. Tissue regions are annotated by combining histological features (from the matched H&E or IF image) with molecular signatures. For example, necrotic regions, fibrotic zones, or lymphoid aggregates are identified by their histological appearance and confirmed by expression of relevant marker genes.
Spatial trajectory and gradient analysis. For tissues with continuous spatial gradients (e.g., developmental axes, wound healing gradients, tumor-host interfaces), spatial trajectory inference maps gene expression along the gradient direction, identifying genes with progressive expression changes across tissue space.
Cell-cell proximity and colocalization. For single-cell-resolution platforms (Xenium, CosMx), nearest-neighbor distances between cell types are computed to quantify spatial colocalization, exclusion, or infiltration patterns. For spot-level data, cell-type spatial correlation matrices infer which cell populations tend to co-occur or segregate in tissue space.
Visualization and Report Delivery
Every analysis result is delivered as a publication-ready figure, annotated with methods, parameters, and interpretation notes.
Standard visualizations
- Spatial feature plots: gene expression heatmaps overlaid on tissue histology images
- Spatial cluster and domain maps: color-coded spatial maps of identified clusters and tissue domains
- UMAP/t-SNE projections: dimensionality reduction plots with cluster, sample, and condition labels
- Spatial variable gene plots: ranked gene lists, spatial expression patterns of top SVGs
- Differential expression visualizations: volcano plots, dot plots, and heatmaps for compartment- and condition-specific comparisons
- Cell-type composition plots: stacked bar charts, spatial pie charts, and proportion heatmaps
- Spatial proximity and interaction maps: nearest-neighbor distance histograms, spatial correlation heatmaps
Interactive HTML reports. Key spatial maps are delivered as interactive HTML reports, allowing exploration of gene expression, cluster composition, and spatial patterns across the tissue section with zoom, pan, and gene search functionality.
Deliverables
Every spatial transcriptomics data analysis project includes structured, ready-to-use deliverables.
- Processed data
- Filtered and normalized expression matrices with spatial coordinates, cell/cluster/domain annotations, and sample metadata (.h5ad, .rds, .h5, .csv)
- QC report
- Per-sample QC metrics and visual summaries, spot/bin/cell filtering documentation, batch correction assessment
- Spatial analysis report
- Clustering and domain identification results, cell-type annotation with marker validation, spatial differential expression results, SVG detection, and region-specific analyses
- Publication-ready figures
- Spatial feature maps, UMAP/t-SNE plots, volcano plots, dot plots, heatmaps, and domain maps (vector PDF/SVG + 300+ dpi TIFF/PNG)
- Reproducible analysis code
- Fully documented R/Python scripts with session information and conda environment files
- Methods write-up
- A draft methods section describing all analytical steps, parameters, software versions, and databases used
Spatial Transcriptomics Data Analysis Applications
Our spatial transcriptomics data analysis service supports a wide range of biological research questions.
Tumor microenvironment characterization
Spatial transcriptomics data analysis maps the transcriptional landscape of tumor, immune, and stromal compartments — identifying spatial gene expression gradients in the tumor microenvironment, immune exclusion patterns, and molecular signatures of tertiary lymphoid structures. See also: Spatial Omics Solutions for Tumor Microenvironment.
Developmental biology and tissue organization
Spatial analysis of developing tissues reveals gene expression gradients that define anatomical axes, tissue layers, and organ primordia. Spatial trajectory analysis maps progressive transcriptional changes along developmental or differentiation paths.
Neuroscience and brain mapping
Spatial transcriptomics data from brain sections is analyzed to identify cortical layer-specific gene expression, regional specialization across brain nuclei, and disease-associated spatial reorganization. See also: Spatial Omics Solutions for Neuroscience.
Drug response and spatial pharmacology
Comparing spatial gene expression between treated and untreated conditions identifies spatially restricted drug effects — revealing which tissue compartments respond to treatment and which are pharmacologically protected.
Multi-sample cohort studies
For cohort studies with multiple samples per condition, our pseudobulk analysis frameworks provide statistically robust comparisons of spatial gene expression across groups, accounting for within-sample spatial heterogeneity.
Case Study: Systematic Benchmarking of Spatially Variable Gene Detection Methods
Source: Chen Y, Zhang J, Yang P, et al., Genome Biology, 2024
Background
Spatially variable gene (SVG) detection — identifying genes whose expression varies systematically across tissue space — is a foundational step in spatial transcriptomics data analysis. Over a dozen computational methods have been proposed for this task (SPARK-X, SpatialDE, nnSVG, SPARK, Trendsceek, and others), but choosing the most appropriate method for a given dataset remains challenging because systematic benchmarking across diverse tissue types, spatial technologies, and resolutions has been lacking.
Methods
A panel of 8 popular SVG detection methods was systematically evaluated on 31 spatial transcriptomics datasets spanning six tissue types (brain, liver, kidney, spleen, heart, lung), three spatial technologies (10x Visium, Slide-seqV2, Stereo-seq), and multiple spatial resolutions. Each method was assessed across five evaluation criteria: statistical calibration, spatial pattern detection sensitivity, computational efficiency, robustness to parameter settings, and reproducibility across biological replicates.
Results
No single method performed best across all criteria. SPARK-X and nnSVG demonstrated the best balance of statistical calibration and sensitivity for 10x Visium data. SpatialDE was most sensitive for detecting genes with gradual spatial gradients but showed inflated false positive rates. Memory usage varied over two orders of magnitude — from ~200 MB to >45 GB per dataset. Importantly, method ranking shifted substantially depending on tissue type and spatial resolution, demonstrating that platform- and tissue-specific benchmarking is essential for selecting appropriate analytical tools. All benchmarking results and reusable analysis scripts were released publicly through the SVGbench framework.
Conclusion
This study exemplifies the rigorous, multi-method evaluation approach that CD Genomics applies when selecting analytical tools for spatial transcriptomics projects. Rather than defaulting to a single SVG detection method, we assess method suitability against dataset characteristics — spatial resolution, tissue type, and sample structure — ensuring that the resulting SVG calls are robust and reproducible. The benchmarking framework and analytical standards demonstrated in this study directly inform the spatial differential expression analysis component of our service.
Adapted from Chen et al., Genome Biology, 2024, doi:10.1186/s13059-023-03145-y, Figure 2.
Frequently Asked Questions (FAQ)
References
- Chen Y, Zhang J, Yang P, et al. "Evaluating spatially variable gene detection methods for spatial transcriptomics data." Genome Biology, vol. 25, 2024, 18.
- Palla G, Fischer DS, Regev A, Theis FJ. "Spatial components of molecular tissue biology." Nature Biotechnology, vol. 40, 2022, pp. 308–318.
- He S, Bhatt R, Brown C, et al. "High-plex imaging of RNA and proteins at subcellular resolution in fixed tissue by spatial molecular imaging." Nature Biotechnology, vol. 40, 2022, pp. 1794–1806.
- Longo SK, Guo MG, Ji AL, Khavari PA. "Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics." Nature Reviews Genetics, vol. 22, 2021, pp. 627–644.
- Hao Y, Hao S, Andersen-Nissen E, et al. "Integrated analysis of multimodal single-cell data." Cell, vol. 184, no. 13, 2021, pp. 3573–3587.e29.
- Moses L, Pachter L. "Museum of spatial transcriptomics." Nature Methods, vol. 19, 2022, pp. 534–546.
