Sample Submission Guidelines
What Is AI-Assisted Spatial Omics Analysis
Spatial omics analysis measures gene or protein expression while preserving each measurement's physical location within a tissue section. Unlike dissociated single-cell or bulk sequencing, spatial data keeps the tissue map intact, so cell types, gene programs, and cell-cell interactions can be studied in their native architecture rather than in isolation.
Why Spatial Data Needs AI-Assisted Analysis
Most spot-based spatial platforms do not resolve single cells directly: each measurement spot can contain signal from several cells of different types, and image-based platforms generate very large, noisy datasets that are impractical to annotate by hand.
- Deconvolution: Spot-level spatial data is a mixture of cell types, so a reference-informed computational model is needed to estimate the cell-type composition of each spot.
- Scale: A single spatial section can contain tens of thousands of spots or millions of segmented cells, well beyond what manual review can consistently annotate.
Note: We select the deconvolution or domain-detection method based on the platform's resolution and the research question, rather than applying one default pipeline to every project.
Spot-based platforms (Visium, Visium HD) require deconvolution to resolve cell-type mixtures; single-cell-resolution platforms (Xenium, CosMx, Stereo-seq) support direct cell typing.
Supported Spatial Platforms and Data Inputs
- Accepted input: FFPE or fresh-frozen tissue sections, or raw/processed Visium output files
- Resolution: spot-based (55 µm); requires deconvolution for cell-type composition
- Accepted input: FFPE tissue sections, or raw/processed Visium HD output files
- Resolution: near single-cell (2 µm bins); still benefits from deconvolution at finer bin sizes
- Accepted input: fixed tissue sections, or processed Xenium transcript and cell-boundary output
- Resolution: subcellular, targeted gene panel; supports direct single-cell typing
- Accepted input: fresh-frozen tissue sections, or processed Stereo-seq expression matrices
- Resolution: subcellular, whole-transcriptome; large field-of-view tissue mapping
Project Entry Options: Data-to-Insight and Sample-to-Insight
Two ways to start a spatial omics project
Data-to-Insight
- You provide existing spatial output files (raw reads, count matrices, or processed platform output) from any supported spatial platform
- We handle QC, deconvolution or cell typing, spatial domain detection, and interpretation
- Best for labs with spatial data already generated in-house or from a prior vendor
Sample-to-Insight
- You provide FFPE or fresh-frozen tissue sections; qualified partner platforms generate the spatial data
- Data then flows into the same AI-assisted spatial analysis pipeline
- Best for teams that need one coordinated project spanning tissue processing and spatial analysis
AI-Assisted Spatial Omics Analysis Workflow
Our workflow connects tissue or data intake, per-section QC, AI-assisted deconvolution or cell typing, and spatial interpretation into one project plan, with QC checkpoints at every stage.
Step 1: Project Start
Project discussion to confirm the spatial platform, project entry option, reference single-cell data availability, and analysis plan.
Step 2: Tissue / Data Reception & QC
Tissue section quality review for Sample-to-Insight projects; file format and image/matrix QC for submitted spatial data.
Step 3: Deconvolution or Cell Typing
Reference-informed deconvolution for spot-based platforms, or direct cell segmentation and typing for single-cell-resolution platforms.
Step 4: Spatial Domain & Neighborhood Analysis
Spatial domain detection, cell-neighborhood enrichment, and, where relevant, ligand-receptor or cell-cell interaction inference.
Step 5: Interpretation & Delivery
Biological interpretation of spatial patterns, delivery of the annotated spatial map and analysis report, and reproducible code.
Choosing the Right Spatial Analysis Method
Method choice depends on platform resolution, tissue type, and the research question. We select and combine methods accordingly rather than defaulting to a single pipeline.
| Analysis Task | Question It Answers | Best-Fit Platform Type | Method Family |
|---|---|---|---|
| Deconvolution | Which cell types make up each spot? | Spot-based (Visium, Visium HD) | Reference-informed decomposition (e.g., RCTD-style models) |
| Direct Cell Typing | What cell type is this individual cell? | Single-cell-resolution (Xenium, CosMx, Stereo-seq) | Segmentation plus reference-based or marker-based classification |
| Spatial Domain Detection | What distinct tissue regions exist, and where are their borders? | Any platform with spatial coordinates | Spatially aware clustering (graph- or neighborhood-informed) |
| Cell-Neighborhood & Interaction Analysis | Which cell types co-occur, and what signaling may be occurring locally? | Single-cell-resolution or fine-bin spot data | Neighborhood enrichment and ligand-receptor colocalization models |
Validation and Quality Control
Spatial results are only as trustworthy as the validation behind them. Every project includes explicit checks for the failure modes most common in spatial analysis.
- Tissue and image quality: Section integrity, staining quality, and read or transcript depth per spot or cell are reviewed before downstream analysis
- Reference mismatch: Deconvolution accuracy depends on how well the single-cell reference matches the tissue and condition; reference suitability is assessed before use
- Segmentation and doublet risk: Cell segmentation quality and doublet or spillover risk are reviewed for single-cell-resolution platforms before cell typing
Applications of Spatial Omics Analysis in Research
Spatial omics analysis supports a wide range of research questions where tissue architecture matters to the biology.
Tumor Microenvironment Mapping
Resolve immune, stromal, and malignant cell distribution and their spatial relationships within tumor tissue.
Cell-Neighborhood and Niche Discovery
Identify recurring multicellular neighborhoods and niches associated with disease state or treatment response.
Single-Cell and Spatial Integration
Combine single-cell reference data with spatial coordinates to map cell states onto their native tissue location.
Developmental and Neuroscience Tissue Mapping
Chart spatial gene expression gradients and regional cell-type organization across developing or neural tissue.
Spatially Resolved Biomarker Discovery
Identify candidate markers whose expression or spatial pattern differs between disease regions and adjacent normal tissue.
Deliverables
- Per-section QC report
- Cell-type deconvolution or cell-typing table with spatial coordinates
- Spatial domain map and cluster annotation
- Cell-neighborhood interaction and, where applicable, ligand-receptor colocalization summary
- Reproducible analysis code and environment files
- Methods-ready write-up describing the analysis approach and validation strategy
Sample and Data Requirements
Minimum requirements vary by platform and tissue type; the table below summarizes typical starting points.
| Platform | Accepted Input | Typical Minimum | Recommended Metadata |
|---|---|---|---|
| 10x Visium / Visium HD | FFPE or fresh-frozen sections, or raw/processed output files | Intact section with RIN/DV200 meeting platform QC thresholds | Tissue type, fixation method, section thickness |
| 10x Xenium In Situ | Fixed tissue sections, or processed transcript/cell-boundary output | Section area within instrument field-of-view limits | Gene panel used, tissue type, fixation protocol |
| Stereo-seq | Fresh-frozen sections, or processed expression matrices | Section area and RNA integrity meeting platform QC thresholds | Tissue type, embedding method, section thickness |
| A matched single-cell RNA-seq reference (existing or newly generated) substantially improves deconvolution accuracy for spot-based platforms; discuss availability with your project manager. | |||
Study Design Requirements
- Representative tissue sections that capture the region or structure relevant to the research question
- Adequate biological replicates per group; single sections limit generalizable conclusions about group differences
- A matched or well-suited single-cell reference dataset when deconvolution is required
- Clearly defined regions of interest or annotation criteria if region-specific comparison is planned
- Consistent tissue processing and staining protocols across samples to limit batch-driven artifacts
Limitations
- Spot-based platforms do not measure true single-cell resolution; deconvolution results are estimates, not direct measurements
- Deconvolution accuracy depends on how well the single-cell reference represents the tissue and condition studied
- Single sections capture one plane of a three-dimensional tissue and may not represent the whole structure
- Spatial correlation between cell types does not establish direct cell-cell interaction; findings support hypothesis generation and should be confirmed with orthogonal validation
Reference
- Longo, Sophia K., et al. "Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics." Nature Reviews Genetics 22.10 (2021): 627-644. https://doi.org/10.1038/s41576-021-00370-8
- Moses, Lambda, and Lior Pachter. "Museum of spatial transcriptomics." Nature Methods 19.5 (2022): 534-546. https://doi.org/10.1038/s41592-022-01409-2
- Cable, Dylan M., et al. "Robust decomposition of cell type mixtures in spatial transcriptomics." Nature Biotechnology 40.4 (2022): 517-526. https://doi.org/10.1038/s41587-021-00830-w
- Walsh, Ian, et al. "DOME: recommendations for supervised machine learning validation in biology." Nature Methods 18.10 (2021): 1122-1127. https://doi.org/10.1038/s41592-021-01205-4
Demo Results
Spatial Deconvolution of Immune and Stromal Cell Types in Tumor Tissue (Ge Q et al., npj Digit Med, 2026)
Cell-Neighborhood Map of an Immune-Suppressive Tumor Niche (Ge Q et al., npj Digit Med, 2026)
Explainable Model Feature Attribution for a Spatially Informed Risk Score (Ge Q et al., npj Digit Med, 2026)
References
- Ge Q, Wang Z, Wang Y, et al. Integrative single-cell and spatial transcriptomics with explainable AI reveal lethal prognostic axis in prostate cancer. npj Digit Med. 2026;9(1):119. https://doi.org/10.1038/s41746-025-02297-4
AI-Assisted Spatial Omics Analysis FAQs
1. Which spatial platform is right for my project?
It depends on whether you need whole-transcriptome coverage or a targeted gene panel, and whether single-cell resolution matters for your question. Spot-based platforms like Visium give broad transcriptome coverage but need deconvolution; single-cell-resolution platforms like Xenium or Stereo-seq support direct cell typing but may use a smaller gene panel. We help match the platform to your research question during project planning.
2. Do I need a single-cell RNA-seq reference for deconvolution?
For spot-based platforms, a matched or well-suited single-cell reference substantially improves deconvolution accuracy. If you do not have one, we can advise on suitable public references or discuss generating a matched reference as part of the project.
3. Can you start from tissue sections instead of finished spatial data?
Yes. Under the Sample-to-Insight option, you provide FFPE or fresh-frozen tissue sections, and qualified partner platforms generate the spatial data, which then flows into the same AI-assisted analysis pipeline.
4. How many tissue sections or replicates do I need?
This depends on your research question. Single sections can support exploratory analysis, but comparing groups or drawing generalizable conclusions typically requires multiple biological replicates per group; we discuss the right design during project planning.
5. Can spatial analysis tell me which cells are physically interacting?
Spatial proximity and neighborhood enrichment can identify cell types that co-occur and may be interacting, and ligand-receptor colocalization can suggest candidate signaling pairs. These results indicate spatial association, not direct proof of interaction, and typically support further experimental validation.
AI-Assisted Spatial Omics Analysis Case Study
Independent Research Example
This publication is not a CD Genomics customer project.
Integrative single-cell and spatial transcriptomics with explainable AI reveal lethal prognostic axis in prostate cancer
Journal: npj Digital Medicine
Published: 6 January 2026
Background
Prostate cancer is clinically heterogeneous, and it remains difficult to translate molecular heterogeneity into a prognostic tool that is both accurate and biologically interpretable. This study integrated single-cell and spatial transcriptomics with explainable machine learning to define a lethal tumor axis and build an interpretable prognostic model for prostate cancer.
Materials & Methods
Cohort
- 141,986 single cells
- Localized, hormone-sensitive, and castration-resistant prostate cancer
- Matched spatial transcriptomic sections
Spatial Analysis
- RCTD-based cell-type deconvolution
- MISTy spatial dependency modeling
- Cell-neighborhood and niche mapping
- 101 machine learning pipelines benchmarked
- Lasso + PLS-Cox prognostic model
- SHAP explainability analysis
Results
- A Malignant Epithelial Subpopulation Drives the Lethal Axis
- A malignant C4 epithelial subpopulation was identified, characterized by high chromosomal instability, androgen receptor and cell-cycle activation, and stemness potential.
- Spatial Mapping Reveals Immune-Suppressive Niches
- Spatial mapping revealed immune-enriched yet suppressive niches, where fibroblasts and myeloid cells coexisted with exhausted lymphocytes, reflecting functional immune imbalance.
- An Explainable Model Delivers a Validated Prognostic Score
- Among 101 benchmarked pipelines, a Lasso plus PLS-Cox model achieved strong concordance across independent cohorts.
- The C4-based risk score independently predicted recurrence-free survival after adjustment for age, Gleason score, and T stage.
- SHAP interpretation highlighted MT1M, PCSK1N, and ACSL3 as the major risk-driving features.
- Experimental Follow-Up Supported the Top Feature
- PCSK1N was progressively upregulated from normal prostate to castration-resistant disease and promoted proliferation, clonogenicity, and migration.
- PCSK1N inhibition sensitized organoids and xenografts to AR-targeted therapy.
Conclusion
Combining spatial deconvolution and neighborhood mapping with an explainable machine learning framework identified a biologically coherent, spatially grounded prognostic signature in prostate cancer, and the top computational feature held up under independent experimental testing. The study illustrates how AI-assisted spatial omics analysis can move from a tissue map to a validated, interpretable research finding.
Reference
- Ge, Qintao, et al. "Integrative single-cell and spatial transcriptomics with explainable AI reveal lethal prognostic axis in prostate cancer." npj Digital Medicine 9.1 (2026): 119. https://doi.org/10.1038/s41746-025-02297-4
Related Publications
Here are some publications that have been successfully published using our services or other related services:
Targeting the CLK2/SRSF9 splicing axis in prostate cancer leads to decreased ARV7 expression
Journal: Molecular Oncology
Year: 2024
Distinct functions of wild-type and R273H mutant Δ133p53α differentially regulate glioblastoma aggressiveness and therapy-induced senescence
Journal: Cell Death & Disease
Year: 2024
High-Fat Diets Fed during Pregnancy Cause Changes to Pancreatic Tissue DNA Methylation and Protein Expression in the Offspring: A Multi-Omics Approach
Journal: International Journal of Molecular Sciences
Year: 2024
The HLA class I immunopeptidomes of AAV capsid proteins
Journal: Frontiers in Immunology
Year: 2023
See more articles published by our clients.
