Spatial Transcriptomics Data Analysis: A Practical Workflow from Raw Data to Biological Insights

Spatial transcriptomics generates genome-wide expression measurements that retain the physical location of each capture spot or cell within an intact tissue section. This workflow article gives researchers, students, analysts, and life-science professionals a reproducible path from raw sequencing output or imaging data through quality control, normalization, clustering, and biological interpretation. The practical outcome is an analysis checklist that supports robust and interpretable results across the major spatial transcriptomics platforms.

Scope and Reader Context

Spatial transcriptomics encompasses two broad technology families. Sequencing-based methods capture mRNA from spatially barcoded spots on a slide, while imaging-based methods use fluorescence in situ hybridization or in situ sequencing to read transcripts directly in tissue. Both families produce large, structured datasets that require specialized computational handling. The workflow described here applies to both, with platform-specific adjustments noted where they matter.

Researchers entering this field typically have experience with bulk RNA-seq or single-cell RNA-seq but face new challenges in spatial data: spot-level sparsity, spatial autocorrelation, tissue morphology alignment, and the need to integrate histological images with molecular measurements. This article assumes familiarity with basic RNA-seq concepts but does not assume prior spatial analysis experience.

At a Glance

Core Principles of Spatial Transcriptomics Analysis

Spatial Context Changes the Analysis Logic

Bulk and single-cell RNA-seq analyses treat each sample or cell as an independent measurement. Spatial transcriptomics adds the tissue coordinate system as a primary analytical dimension. Spots that are physically close tend to share expression programs because they sample adjacent cells or even parts of the same cell. This spatial autocorrelation is biologically meaningful and should be preserved through the analysis instead of treated as a nuisance.

The integration of spatial information with transcriptional profiles enables analyses that are impossible with dissociated cells. Researchers can map cell types back to their tissue niches, identify spatial domains with distinct molecular programs, and infer cell-cell communication based on physical proximity. These capabilities have driven applications in oncology, neurobiology, and developmental biology.

Data Structure and Platform Differences

Sequencing-based platforms such as 10x Genomics Visium capture mRNA from tissue sections placed on slides with spatially barcoded capture probes. Each capture spot is roughly 55 micrometers in diameter and may contain multiple cells. The output is a count matrix where rows are genes and columns are spots, plus a histological image of the stained tissue section.

Imaging-based platforms such as MERFISH, seqFISH, and Xenium read transcripts directly in tissue using combinatorial fluorescence labeling or in situ sequencing. These methods achieve single-cell or subcellular resolution but typically measure a curated gene panel instead of the whole transcriptome. The analytical challenges differ accordingly: imaging data requires cell segmentation, transcript assignment, and barcode decoding before expression matrices can be constructed.

A review of imaging-based spatial transcriptomics describes the analytical framework that converts raw fluorescence signals into molecule-, cell-, and tissue-level representations, with preprocessing, registration, restoration, feature detection, barcode decoding, molecule calling, cell segmentation, transcript assignment, probabilistic cell typing, spatial-domain inference, and atlas integration as the key computational problems. Optical crowding, tissue thickness, panel bias, and multimodal complexity increase the difficulty of these steps.

Reproducibility and Data Management

Spatial transcriptomics datasets are large and complex. Reproducible analysis requires version control for code, documented software environments, and systematic record keeping for parameters and quality metrics. The FAIR Guiding Principles provide a framework for making data findable, accessible, interoperable, and reusable. Applying these principles to spatial transcriptomics means using standard file formats, depositing processed data in public repositories, and documenting analysis steps thoroughly.

The National Institutes of Health Genomic Data Sharing Policy establishes expectations for data sharing in NIH-funded research. Researchers should review these requirements early in their project planning to ensure compliance. The NCBI Data Resources and EMBL-EBI Training offer guidance on data deposition and analysis best practices.

Raw Data Processing

Sequencing-Based Platforms

Raw sequencing data from spatial transcriptomics experiments arrive as FASTQ files containing reads from spatially barcoded capture probes. The first processing step is to demultiplex reads by spatial barcode, map them to the reference genome, and generate a count matrix. Platform vendor pipelines handle this for their own chemistries, but open-source alternatives offer flexibility and scalability.

Spacemake is a modular spatial transcriptomics pipeline built in Snakemake and Python that handles all major spatial transcriptomics data types and can be configured for other technologies. It processes and analyzes multiple samples in parallel, even when samples come from different experimental methods. Spacemake provides a unified framework for reproducible processing from raw sequencing data to automatically generated downstream analysis reports, with additional modules for sample merging, saturation analysis, and long-read analysis. It also integrates spatial and single-cell transcriptomics data using novoSpaRc to increase gene counts for the spatial dataset.

For researchers working with large datasets, pipeline scalability matters. Spacemake is designed to handle large-scale data and can be integrated with existing computational workflows. The choice between vendor pipelines and open-source alternatives depends on the platform, the scale of the study, and the need for custom processing steps.

Imaging-Based Platforms

Imaging-based spatial transcriptomics requires a different raw processing path. Raw fluorescence images must be registered to a common coordinate system, corrected for optical distortions, and processed to detect individual transcripts. Barcode decoding assigns each detected transcript to its target gene. Cell segmentation defines cell boundaries so that transcripts can be assigned to cells.

These steps are computationally intensive and error-prone. Optical crowding, where many transcripts are packed into a small area, can cause missed detections or misassignments. Tissue thickness affects signal quality and background levels. Panel bias, where some genes are measured more reliably than others, must be accounted for in downstream analysis.

The review of imaging-based spatial transcriptomics emphasizes that despite major differences in chemistry, coding, and imaging strategies across platforms, biological interpretation converges on a common set of computational problems. Researchers should expect to spend substantial effort on preprocessing quality before any biological analysis begins.

Long-Read and Emerging Technologies

Some spatial transcriptomics protocols generate long-read data that can capture full-length transcripts and isoform information. Spacemake includes a module for long-read analysis. Emerging technologies continue to expand the spatial profiling landscape, and pipelines must adapt to new data formats and chemistries.

Quality Control

Spot-Level Quality Metrics

Quality control for spatial transcriptomics operates at multiple levels. At the spot level, key metrics include total read count, number of detected genes, and fraction of reads mapping to mitochondrial genes. Low-quality spots typically have low read counts, few detected genes, and high mitochondrial fractions, indicating damaged or dying cells.

For imaging-based data, quality metrics include the number of transcripts detected per cell, the fraction of cells with sufficient transcript coverage, and the accuracy of cell segmentation. Cells with very low transcript counts may be segmentation artifacts or dying cells and should be flagged for review.

Tissue Coverage and Histology Alignment

Spatial transcriptomics data must be aligned with the histological image of the tissue section. Misalignment between the molecular data and the image can cause incorrect assignment of spots to anatomical regions. Researchers should visually inspect the alignment for each sample and document any corrections applied.

The Danish SURVIVE study protocol for analyzing RNA expression in long-term formalin-fixed human hypothalamus samples demonstrates the importance of tissue quality control. The study used baking at 60 degrees Celsius for 1 hour followed by 30 minutes of heat-induced target retrieval at 99 degrees Celsius to achieve optimal RNA availability while maintaining tissue integrity. This protocol enabled whole-transcriptome analysis of microglia and neuron populations using the GeoMx Digital Spatial Profiler, with the StandR tool used for quality control and data analysis.

Batch Effects and Technical Variation

Spatial transcriptomics experiments often involve multiple tissue sections, slides, or sequencing runs. Technical variation between batches can confound biological comparisons. Researchers should record batch information for every sample and test for batch effects before proceeding with downstream analysis.

The ssREAD database for Alzheimer's disease provides an example of large-scale data integration. It encompasses 1,053 samples from 277 integrated datasets totaling over 7 million cells from single-cell and single-nucleus studies, plus 381 spatial transcriptomics datasets from human and mouse brain studies. Each dataset is annotated with species, gender, brain region, disease or control status, age, and Braak stage. This level of metadata annotation is essential for cross-dataset analysis and for identifying batch effects.

Normalization

Library Size Normalization

Spatial transcriptomics spots vary in sequencing depth due to differences in tissue capture efficiency, RNA quality, and sequencing allocation. Library size normalization scales each spot's counts to a common total, typically the median library size across spots. This adjustment is essential before any cross-spot comparison.

Log transformation is commonly applied after normalization to stabilize variance and make the data more amenable to downstream statistical methods. Some methods use variance-stabilizing transformations that account for the mean-variance relationship in count data.

Platform-Specific Considerations

Imaging-based spatial transcriptomics data have different noise characteristics than sequencing-based data. Transcript detection is limited by the gene panel and by optical crowding effects. Normalization must account for panel bias and for differences in detection efficiency across genes.

The ALOA pipeline for spatial profiling imaging data provides an example of platform-specific analysis. ALOA is designed to handle spatial coordinates and image-based data from multiplexed immunohistochemistry and immunofluorescence, providing analysis with accompanying graphs and statistical output. It is not suited for single-cell transcriptomics or non-spatial data, which require tools designed for high-dimensional gene expression information.

When Normalization Is Not Enough

Some spatial datasets show spatial gradients in technical quality that cannot be fully corrected by global normalization. For example, tissue edges may have lower capture efficiency than the center. Researchers should examine spatial patterns in quality metrics and consider whether apparent biological gradients are technical artifacts.

Dimensionality Reduction and Clustering

Principal Component Analysis

Principal component analysis (PCA) is the standard first step for reducing the dimensionality of spatial transcriptomics data. PCA identifies the major axes of variation in the normalized expression matrix, compressing thousands of genes into a few dozen principal components that capture most of the biological signal.

The choice of how many principal components to retain affects downstream clustering. Too few components lose biological signal, while too many introduce noise. Researchers typically examine the variance explained by each component and select a cutoff based on the elbow in the variance curve or on the number of expected cell types or spatial domains.

Clustering Methods for Spatial Data

Clustering assigns spots to groups based on transcriptional similarity. Standard single-cell clustering methods can be applied to spatial data, but they ignore spatial information. Spatial clustering methods incorporate spot coordinates to encourage neighboring spots to cluster together, producing more coherent spatial domains.

A comparison of spatial transcriptomics clustering methods applied BayesSpace, k-means, and Gaussian mixture models to melanoma data. BayesSpace, which incorporates spatial information into the clustering model, produced the most accurate results. Marker gene staining verified the clustering output and enabled precise identification of cell types.

Graph-based methods have become popular for spatial clustering. ADEPT uses a graph autoencoder backbone with iterative clustering on imputed, differentially expressed gene matrices to minimize variance in clustering results. It outperformed other methods on spatial domain identification, visualization, spatial trajectory inference, and data denoising across different platforms.

Marker gene information can improve clustering. MGGNN, a marker gene-guided graph neural network approach, trains a model using contrastive learning and then fine-tunes it using spots labeled by marker gene expression. Simulation and experiments on real datasets demonstrated superior performance over state-of-the-art methods.

Rare cell type detection remains a challenge. aKNNO uses an adaptive k-nearest neighbor graph with optimization to simultaneously identify abundant and rare cell types. Benchmarking on 38 simulated and 20 single-cell and spatial transcriptomics datasets showed that aKNNO identifies both abundant and rare cell types more accurately than general and specialized methods.

Spatial Domain Identification

Spatial domains are contiguous regions of tissue with distinct molecular programs. Identifying these domains is a primary goal of spatial transcriptomics analysis. Clustering methods that incorporate spatial information produce domain assignments that can be visualized on the tissue section and compared with histological annotations.

Spotiphy transforms sequencing-based spatial transcriptomics data into single-cell-resolved whole-transcriptome images. It delivers precise cellular proportions in benchmarking evaluations and identifies multiple spatial domains and alterations in tumor-microenvironment interactions in human breast cancer data. Spotiphy also reveals astrocyte and disease-associated microglia regional specifications in Alzheimer's disease and healthy mouse brains.

Cell Type Identification and Deconvolution

Marker-Based Annotation

Once clusters are identified, researchers assign cell type labels based on marker gene expression. This step requires domain knowledge and careful validation. Canonical markers for major cell types are well established, but spatial data may reveal cell states that do not fit neatly into predefined categories.

The muscle-invasive bladder cancer study classified the tumor microenvironment into ten major cell subsets based on canonical markers after quality control. This study used single-cell RNA sequencing and spatial transcriptomics to investigate immune cell subsets, intercellular communication networks, and spatial distributions associated with neoadjuvant immunochemotherapy response.

Deconvolution of Multi-Cell Spots

Sequencing-based spatial transcriptomics spots often contain multiple cells. Deconvolution methods estimate the cell type composition of each spot by integrating single-cell RNA-seq reference data. This integration enables cell type mapping at sub-spot resolution.

The pancreatic carcinogenesis study introduced an imaging, spatial transcriptomics, and single-cell RNA-sequencing integration pipeline to characterize neoplastic cell state transitions. The semi-supervised analysis pipeline examined premalignant pancreatic intraepithelial neoplasias that can develop into pancreatic ductal adenocarcinoma. The study demonstrated that cancer-associated fibroblasts, including antigen-presenting CAFs, are located close to PanINs and observed a transition from CAF-related inflammatory signaling to cellular proliferation during PanIN progression.

Probabilistic Cell Typing

Imaging-based spatial transcriptomics data can be analyzed at single-cell resolution, but cell typing is complicated by sparse transcript coverage. Probabilistic cell typing methods assign each cell a probability distribution over cell types instead of a single label. This approach accounts for uncertainty and enables more nuanced downstream analysis.

Spatial Analysis and Biological Interpretation

Spatially Variable Gene Detection

Spatially variable genes show expression patterns that vary across the tissue in a non-random way. Identifying these genes is a key step in characterizing tissue organization. Methods for spatial variable gene detection use spatial autocorrelation statistics or model-based approaches to distinguish true spatial patterns from technical noise.

Cell-Cell Communication Analysis

Spatial transcriptomics enables inference of cell-cell communication based on physical proximity and ligand-receptor expression. Cells that are close together and express complementary ligand-receptor pairs are candidates for active signaling. The dry age-related macular degeneration study integrated spatial transcriptomics and single-cell RNA sequencing to identify an endothelial-macrophage signaling pathway driving disease progression. The study found that endothelial cells regulate SLC16A10-positive macrophages through the TNFSF10-TNFRSF10B pathway, with NFKB1 activating NF-kB signaling and promoting formation of a vascular-immune inflammatory niche.

Trajectory and Transition Analysis

Spatial data can reveal transitions in cell state across tissue regions. Pseudotime trajectory analysis orders cells or spots along a developmental or disease progression axis. The pancreatic carcinogenesis study used this approach to track progression from low-grade to high-grade PanIN lesions and map cellular phenotypes relative to single-cell pancreatic ductal adenocarcinoma datasets.

Integration with Other Data Modalities

Spatial transcriptomics is most powerful when integrated with other data types. The head and neck squamous cell carcinoma study integrated transcriptomic data from multiple GEO datasets and the TCGA-HNSC cohort, used four machine learning algorithms to screen for core candidate genes, and dissected cellular origins and spatial distribution using public single-cell and spatial transcriptomics data. The study found that SASH1 was specifically downregulated within the malignant cell population and its expression was spatially exclusive from COL1A1-high fibrotic stromal regions.

A computational framework for spatial mechano-transcriptomics enables joint statistical analysis of transcriptional and mechanical signals. Applied to developing mouse embryo data, the framework infers forces acting on individual cells and identifies mechanical, morphometric, and gene expression signatures predictive of tissue compartment boundaries. Geoadditive structural equation modeling identifies gene modules that predict mechanical behavior of cells.

Practical Implementation Steps

Step 1: Define the Analysis Plan

Before processing any data, write an analysis plan that specifies the research question, the expected outputs, the software environment, and the quality thresholds. This plan should be version-controlled and shared with collaborators.

Step 2: Set Up the Computational Environment

Spatial transcriptomics analysis requires substantial computational resources. Most steps can run on a standard workstation for small datasets, but large studies may require high-performance computing. Document the software versions and dependencies for every tool used.

Step 3: Process Raw Data

Run the platform-specific pipeline to generate count matrices and spatial coordinates. For sequencing-based data, use the vendor pipeline or an open-source alternative such as Spacemake. For imaging-based data, complete image registration, transcript detection, barcode decoding, and cell segmentation.

Step 4: Perform Quality Control

Generate quality control reports for every sample. Record spot-level metrics, examine spatial patterns in quality, and check histology alignment. Apply filtering thresholds and document the number of spots and genes retained.

Step 5: Normalize and Transform

Apply library size normalization and log transformation. Record the normalization method and parameters. Check that normalization successfully removed technical variation without removing biological signal.

Step 6: Reduce Dimensionality and Cluster

Run PCA and select the number of principal components. Apply clustering methods appropriate for the data type and research question. Compare results from multiple methods when possible.

Step 7: Annotate and Interpret

Assign cell type labels based on marker genes. Identify spatial domains and spatially variable genes. Perform downstream analyses such as deconvolution, cell-cell communication, and trajectory inference as appropriate for the research question.

Step 8: Document and Report

Record all parameters, quality metrics, and analysis decisions. Generate visualizations of the spatial data and clustering results. Prepare a reproducibility report that allows another researcher to rerun the analysis.

Records and Measurements

Essential Records for Every Analysis

Quality Metrics to Monitor

Track the number of spots passing quality filters, the median number of genes detected per spot, the fraction of reads mapping to the genome, and the fraction of reads mapping to mitochondrial genes. For imaging-based data, track the number of transcripts detected per cell and the fraction of cells passing quality thresholds.

Common Failure Patterns

Overfiltering

Aggressive quality filtering can remove biologically interesting spots, particularly in regions with low RNA content such as adipose tissue or dense stroma. Balance the need for clean data against the risk of losing biological signal. Record filtering decisions and consider sensitivity analyses with different thresholds.

Ignoring Spatial Autocorrelation

Standard single-cell analysis methods that ignore spatial information can produce fragmented clusters that do not correspond to coherent tissue domains. Use spatial clustering methods when the research question involves tissue organization.

Misalignment Between Molecular Data and Histology

Poor alignment between the count matrix and the histological image leads to incorrect assignment of spots to anatomical regions. Always visually inspect alignment and document any corrections.

Batch Effects Confounded with Biology

When samples from different conditions are processed in different batches, batch effects can be mistaken for biological differences. Design experiments to balance conditions across batches and test for batch effects explicitly.

Overinterpretation of Deconvolution Results

Deconvolution methods provide estimates, not ground truth. Validate deconvolution results with independent methods such as immunohistochemistry or imaging-based spatial transcriptomics when possible.

Computational Resource Bottlenecks

Spatial transcriptomics datasets are large, and some analysis steps are computationally intensive. Plan for adequate storage and compute resources. Use scalable pipelines such as Spacemake for large studies.

Limitations and Interpretation Boundaries

Resolution Limits

Sequencing-based spatial transcriptomics spots contain multiple cells, limiting the resolution of cell type mapping. Deconvolution methods provide estimates but cannot fully resolve sub-spot heterogeneity. Imaging-based methods achieve single-cell resolution but are limited to gene panels.

Gene Panel Bias

Imaging-based spatial transcriptomics measures a curated set of genes. Genes not in the panel cannot be detected, and panel composition affects the ability to distinguish cell types and states. Results are conditional on the panel design.

Technical Artifacts

Fixation, tissue processing, and storage affect RNA quality and detection. The study of long-term formalin-fixed human brain tissue demonstrated that optimized target retrieval can enable whole-transcriptome analysis, but not all tissue samples will yield usable data. Researchers should assess RNA quality before committing to spatial transcriptomics experiments.

Cross-Platform Comparability

Different spatial transcriptomics platforms produce data with different characteristics. Comparing results across platforms requires careful normalization and validation. The review of imaging-based spatial transcriptomics notes that multimodal complexity increases computational difficulty.

Statistical Power

Spatial transcriptomics experiments often have few biological replicates due to cost and complexity. Limited replication reduces statistical power and increases the risk of false discoveries. Researchers should interpret findings cautiously and validate key results with independent methods.

Safety and Regulatory Context

Data Sharing and Privacy

Spatial transcriptomics data from human samples may contain sensitive information. Researchers must comply with applicable regulations and institutional policies for data handling and sharing. The NIH Genomic Data Sharing Policy establishes expectations for data sharing in NIH-funded research. The NCBI Data Resources provide repositories for depositing and accessing genomic data.

Ethical Use of Data

Spatial transcriptomics data from human tissues should be used in accordance with ethical principles for human subjects research. Researchers should ensure that samples were collected with appropriate consent and that data use is consistent with the consent terms.

Reproducibility Standards

Funding agencies and journals increasingly require data and code availability. Following the FAIR Guiding Principles for data management and stewardship supports reproducibility and maximizes the value of research investments. The EMBL-EBI Training offers resources for researchers learning data management and analysis skills.

Professional Escalation Criteria

When to Seek Expert Help

Consult a bioinformatics specialist or spatial transcriptomics expert when any of the following situations arise:

• Quality control metrics show unexpected patterns that cannot be explained by standard troubleshooting
• Clustering results are unstable across runs or methods
• Deconvolution results conflict with histological observations
• Integration of spatial data with single-cell or other modalities produces inconsistent results
• The analysis requires custom methods not available in standard tools

When to Reconsider the Experimental Design

Some problems cannot be fixed computationally. If tissue quality is poor, if the gene panel is inadequate for the research question, or if the experimental design confounds key variables, new data may be needed. Discuss these issues with collaborators early instead of after extensive analysis.

When to Validate with Independent Methods

Key findings should be validated with independent methods. If spatial transcriptomics identifies a novel cell population, a new spatial domain, or a candidate cell-cell interaction, confirm the finding with immunohistochemistry, imaging-based spatial transcriptomics, or functional experiments before drawing strong conclusions.

Frequently Asked Questions

What is the difference between sequencing-based and imaging-based spatial transcriptomics?

Sequencing-based methods capture mRNA from spatially barcoded spots on a slide and measure the whole transcriptome at spot resolution, typically with multiple cells per spot. Imaging-based methods read transcripts directly in tissue using fluorescence labeling or in situ sequencing, achieving single-cell or subcellular resolution but measuring a curated gene panel. The choice between platforms depends on whether whole-transcriptome coverage or single-cell resolution is more important for the research question.

How many spots or cells should pass quality control?

There is no universal threshold. The number of spots or cells retained depends on the platform, tissue type, and research question. What matters is that the retained data are of sufficient quality for the planned analyses and that filtering decisions are documented and justified. Examine the distribution of quality metrics and choose thresholds that remove clear outliers without eliminating biologically interesting regions.

Which clustering method should I use for spatial transcriptomics data?

The choice depends on the data type and research question. BayesSpace incorporates spatial information and performed well in comparative benchmarking. Graph-based methods such as ADEPT and MGGNN are effective across platforms. aKNNO is designed to identify both abundant and rare cell types. Compare results from multiple methods when possible and validate with marker gene expression.

How do I integrate spatial transcriptomics with single-cell RNA-seq data?

Integration typically involves mapping spatial spots to cell types using single-cell reference data. Deconvolution methods estimate cell type composition for multi-cell spots. Tools such as Spacemake integrate spatial and single-cell data to increase gene counts for spatial datasets. The pancreatic carcinogenesis study provides an example of a semi-supervised integration pipeline applied to FFPE samples.

What are spatially variable genes and why do they matter?

Spatially variable genes show expression patterns that vary across the tissue in a non-random way. They are important because they reveal tissue organization, identify spatial domains, and highlight genes involved in regional biological processes. Methods for detecting spatially variable genes use spatial autocorrelation statistics or model-based approaches.

How do I handle batch effects in spatial transcriptomics data?

Record batch information for every sample and test for batch effects before downstream analysis. Design experiments to balance conditions across batches. If batch effects are detected, apply appropriate correction methods or include batch as a covariate in statistical models. Be cautious about correction methods that may remove biological signal along with technical variation.

Can I analyze spatial transcriptomics data on a standard laptop?

Small datasets can be analyzed on a standard workstation, but large studies require substantial computational resources. The scale of the data, the complexity of the analysis, and the number of samples determine the requirements. Scalable pipelines such as Spacemake are designed for large-scale data and can run on high-performance computing clusters.

How should I validate spatial transcriptomics findings?

Validate key findings with independent methods. Confirm cell type assignments with marker gene staining or immunohistochemistry. Validate deconvolution results with imaging-based spatial transcriptomics or protein-level measurements. The head and neck squamous cell carcinoma study validated SASH1 expression at the protein level via Western blot in cell lines. The melanoma clustering study used marker gene staining to verify clustering results.

Related Bioinformatics Guides

• Spatial Transcriptomics Alignment and Cellular Neighborhood Analysis
• Spatial Transcriptomics: Mapping the Cellular Atlas
• Alternative Splicing Analysis from RNA-Seq Data
• Hidden Markov Models in Biological Sequence Analysis
• GWAS QC Steps: Structural Analysis and Computational Methodologies in Bioinformatics

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• A multi-omics pipeline integrating machine learning and spatial-cellular analysis identifies SASH1 as a prognostic biomarker and therapeutic target in head and neck squamous cell carcinoma.. International journal of surgery (London, England), 2025.
• PanIN and CAF transitions in pancreatic carcinogenesis revealed with spatial data integration.. Cell systems, 2024.
• Analysis of Spatial Molecular Data.. Methods in molecular biology (Clifton, N.J.), 2023.
• A single-cell and spatial RNA-seq database for Alzheimer's disease (ssREAD).. Nature communications, 2024.
• Spacemake: processing and analysis of large-scale spatial transcriptomics data.. GigaScience, 2022.
• A computational pipeline for spatial mechano-transcriptomics.. Nature methods, 2025.
• Spotiphy enables single-cell spatial whole transcriptomics across an entire section.. Nature methods, 2025.
• ALOA, a pipeline for preliminary analysis of spatial profiling imaging data.. Computational and structural biotechnology journal, 2024.
• Mitigating Fixation Artifacts in Spatial Transcriptomics: Methodological Insights from Human Brain Tissue.. 2026.
• Editorial: AI in single-cell biology
• Imaging-Based Spatial Transcriptomics: Data Interpretation Methods and Biomedical Applications.. 2026.
• Remodeling of the tumor microenvironment in muscle-invasive bladder cancer: insights from single-cell and spatial transcriptomics after neoadjuvant immunochemotherapy.. 2026.
• Single-cell and spatial analyses reveal endothelial-macrophage inflammatory crosstalk in dry age-related macular degeneration.. 2026.
• Spatial transcriptomics in Alzheimer's disease: technologies, challenges and discoveries.. 2026.
• A Comparison of Spatial Transcriptomics Clustering Methods Based on Three Different Approaches. Academic Journal of Computing &amp, Information Science, 2024.
• Marker Gene-Guided Graph Neural Networks for Enhanced Spatial Transcriptomics Clustering. AI medicine, 2025.
• aKNNO: single-cell and spatial transcriptomics clustering with an optimized adaptive k-nearest neighbor graph. Genome Biology, 2024.
• ADEPT: Autoencoder with differentially expressed genes and imputation for robust spatial transcriptomics clustering. iScience, 2023.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.