Spatial transcriptomics and spatial epigenomics co-analysis combine spatial gene expression with spatial chromatin readouts from intact tissue. It is used when RNA maps alone are not enough—especially when you need to explain region-specific expression, boundary zones, or microenvironment-driven programs with regulatory evidence such as chromatin accessibility or histone marks.

In real projects, the hardest parts are not "which algorithm is best." They are practical: uneven signal, section-to-section variation, unpaired measurements, and metadata gaps that make alignment fragile. This guide focuses on decisions you can defend in a methods section: study design choices, a workflow that reduces false structure, a compact tool overview, and six published examples that shaped the 2024–2025 landscape.

Figure 1. An end-to-end overview of spatial transcriptomics and spatial epigenomics co-analysis, from study design to deliverables.

When Co-analysis Changes the Biological Story

Co-analysis is most useful when the spatial RNA pattern is clear, but the mechanism is not. Many teams start with a "cell type map," then discover that the interesting biology sits on top of cell type: gradients, niche effects, activation states, or boundary zones that are not neatly explained by marker genes alone.

A Quick Yes/No Checklist

Co-analysis usually earns its cost when you can answer yes to at least two items:

• Yes: A gene program is spatially restricted, but cell types look similar across regions.
• Yes: You need enhancer or promoter evidence to support a regulatory hypothesis.
• Yes: You expect layered tissue structure, sharp boundaries, or gradients.
• Yes: You want to prioritize noncoding regions in a microenvironment context.
• Yes: You need evidence that a spatial domain reflects regulation, not batch.

Co-analysis often adds less value when the goal is basic tissue annotation or broad cell-type localization. In those cases, invest first in spatial RNA quality and interpretation, then add an epigenomic layer only when your conclusions need regulatory support.

What "Changed the Story" Looks Like in Outputs

Strong projects show a shift from "pretty maps" to "testable claims." Look for deliverables like:

• Spatial domains supported by both RNA patterns and chromatin evidence.
• Ranked enhancer–gene hypotheses tied to specific tissue regions.
• Motif activity or regulatory program summaries that match expression changes.
• A short list of region-specific candidates for follow-up perturbation or validation.

Why Spatial Modalities Fail to Line Up

Spatial modalities fail to line up for predictable reasons. Most alignment failures start upstream of modeling: inconsistent section handling, mismatched resolution, low signal in one modality, or missing metadata that makes tissue context ambiguous.

Four Common Friction Points

• Sparsity is different across modalities. Chromatin features are often sparser and more variable than RNA counts.
• Resolution is not automatically compatible. Spots, pixels, and segmented cells represent different statistical units.
• Batch effects can masquerade as domains. Section day, slide, and processing differences can dominate embeddings.
• Unpaired measurements are common. Many studies profile RNA and chromatin in different sections.

Symptom–Cause–Fix Table

Paired vs Unpaired, Defined Once

Paired means both modalities share a coordinate system by design (same section or matched barcodes).

Unpaired means modalities come from different sections or datasets and alignment is inferred.

Unpaired spatial multi-omics integration can still be credible, but it requires stronger validation and more conservative interpretation—especially when imputation is involved.

Common Study Designs: Paired, Unpaired, and Single-Cell-to-Space

Study design determines what you can claim. It also determines which tools and validation steps are appropriate.

Figure 2. Three common study designs for spatial transcriptomics–epigenomics co-analysis: paired, unpaired, and single-cell-to-space mapping.

Three Common Setups

Practical Design Notes That Prevent Rework

• Plan biological replicates early. Section-to-section variability is real, even in well-controlled runs.
• Keep sectioning consistent across conditions. Small changes in section order or thickness can create avoidable shifts.
• Treat metadata as part of the dataset. Slide IDs, section order, staining, and processing dates should travel with the data.
• Decide your unit of analysis up front. If you want cell-level summaries, plan segmentation and validation early.

If high-resolution spatial RNA is your anchor modality, CD Genomics provides a sequencing option on spatial-omicslab.com: 10 µm Spatial Transcriptomics Sequencing Service. In integrated studies, an anchor modality with a stable spatial structure often reduces downstream alignment ambiguity.

Practical Workflow: QC, Alignment, and Interpretation

A workflow that "works once" is not enough. The goal is a workflow that produces stable results, documents decisions, and makes it obvious when alignment is being driven by technical structure.

Figure 3. A seven-step workflow from QC to interpretation and reporting for spatial multi-omics co-analysis.

Seven Steps That Hold Up in Review

1. Confirm the sample map and metadata. Lock section order, stains, slide IDs, and conditions.
2. Run modality-specific QC. Use different thresholds for RNA and chromatin.
3. Choose features intentionally. Select genes and peaks that match the question, not just variance.
4. Build spatial neighborhoods from coordinates. Treat the tissue grid as primary evidence.
5. Integrate or align modalities. Match methods to paired vs unpaired reality.
6. Interpret with guardrails. Tie domains and programs back to histology and markers.
7. Package results for reporting. Provide figures, tables, and reproducibility notes.

Minimum Metadata to Record

These items prevent avoidable disputes later:

• Platform and chemistry versions for each assay.
• Reference genome build and annotation release.
• Fixation details, section thickness, and permeabilization timing.
• Imaging settings and registration parameters.
• Read depth, mapping rates, duplicate rates, and fragment size profiles.
• Peak calling parameters and any blacklist regions used.

Practical Bench Notes: Teams Often Underestimate

• Image registration matters more than it feels. Small shifts can create false boundaries.
• Do not assume chromatin and RNA have the same "effective resolution." Match the unit of analysis to your coverage.
• Keep processing consistent across conditions. Changing fixation or permeabilization mid-study can introduce hard-to-detect artifacts.
• Always retain a "sanity region." Include a region of known biology (or known negatives) as a reality check.

If you want a deeper explanation of how spatial ATAC-seq chemistry and workflow choices influence signal quality, see Spatial ATAC-seq Experimental Workflow and Principles.

If segmentation is part of your plan, treat it as a model choice, not a formatting step. For method tradeoffs and selection criteria, see Cell Segmentation in Spatial Transcriptomics: How to Choose Between StarDist, Cellpose, CellBin and SCS.

Tool Overview for Spatial Transcriptomics–Epigenomics Integration

Tools are easier to choose when you separate the integration goal from the data constraint. "Domain discovery" is a different task than "cross-modal prediction," and both are different from "single-cell-to-space mapping."

Figure 4. A task-based map of common tools used in spatial transcriptomics and epigenomics integration.

Tool Comparison Table

SWITCH

SWITCH is designed for unpaired spatial multi-omics integration and treats integration and cross-modal translation as a unified modeling task. It is useful when RNA and chromatin are not measured in the same coordinate system, and you still need a shared representation for downstream analyses.

Practical guidance

• Use SWITCH when "unpaired" is not a corner case but the reality of the dataset.
• Treat predicted layers as hypotheses and keep uncertainty visible in reporting.
• Validate stability across seeds and neighborhood definitions before interpreting rare niches.

SpatialGlue

SpatialGlue is a graph neural network approach that emphasizes spatial structure and uses attention mechanisms to fuse information within and across modalities. It is commonly used when the primary question is spatial domain structure rather than prediction.

Practical guidance

• Use SpatialGlue when domain boundaries and tissue architecture are central to the study.
• Report how domain calls change with neighbor parameters and feature selection.
• Tie domain interpretations to histology rather than embeddings alone.

SIMO

SIMO supports probabilistic alignment that places multi-omics single-cell states into spatial context. It fits studies where single-cell data provide deep modality coverage, while spatial data provide the coordinate system.

Practical guidance

• Use SIMO when you have strong single-cell references for the same tissue system.
• Interpret low-confidence regions conservatively and show confidence gradients.
• Avoid converting probability maps into hard labels too early.

bindSC

bindSC is often used as a general multi-omics integration reference point. It is not spatial-first, but it can help you reason about alignment assumptions and feature matching.

Practical guidance

• Use bindSC as a baseline for comparisons and sensitivity checks.
• Avoid presenting it as a dedicated spatial integration solution.

Choosing the Epigenomic Layer: Accessibility vs Histone Marks

Assay choice changes what "regulation" means. Accessibility (spatial ATAC-seq) often supports broad regulatory landscape discovery, while histone marks (spatial CUT&Tag) focus interpretation on specific modifications or protein–DNA contexts.

For a focused, practical comparison, see Spatial Epigenomics Explained: Spatial ATAC-seq vs Spatial CUT&Tag for Real Tissues.

If you want a single provider to handle assay selection, sequencing, and analysis planning, CD Genomics offers Spatial Epigenomics Services, including spatial ATAC-seq and spatial CUT&Tag.

Mini Casebook: 6 Examples

Each case below uses the same template to keep the focus on what transfers to real projects.

Case 1: SWITCH — Unpaired Integration and Cross-Modal Prediction

Problem: Integrate spatial transcriptomics and spatial epigenomics when data are unpaired.

Data Setup: Spatial RNA and spatial chromatin from different sections or datasets.

Core Idea: Learn aligned representations and support cross-modal translation without requiring paired inputs.

Outputs: Joint embedding; spatial domains; predicted layers with uncertainty signals.

Sanity Checks: Domain stability across runs; marker plausibility across modalities; batch association tests.

Key Reference: Li et al. "Integrative Deep Learning of Spatial Multi-Omics with SWITCH." Nature Computational Science (2025).

Case 2: eSpatial — Enhancer Combinations Linked to Spatial Expression

Problem: Explain spatially specific expression using enhancer logic rather than cell type alone.

Data Setup: Spatial expression plus chromatin accessibility information.

Core Idea: Infer region-specific enhancer combinations that encode spatial gene expression patterns.

Outputs: Ranked enhancer sets per region; enhancer–gene hypotheses; motif context for candidate TF programs.

Sanity Checks: Co-localization of accessibility and expression; distance-aware constraints; replicate consistency.

Key Reference: Hong et al. "Divergent Combinations of Enhancers Encode Spatial Gene Expression." Nature Communications (2025).

Case 3: SpatialGlue — Spatial Domains From Multi-Modal Graph Structure

Problem: Define domains when modalities differ in noise and distribution.

Data Setup: Spatial epigenome–transcriptome or other spatial multi-omics combinations.

Core Idea: Use spatial graph structure and attention-based fusion to infer domains supported by both modalities.

Outputs: Domain maps; domain markers; fused representations for downstream testing.

Sanity Checks: Boundary plausibility with anatomy; sensitivity to neighbor parameters; replicate concordance.

Key Reference: Long et al. "Deciphering Spatial Domains from Spatial Multi-Omics with SpatialGlue." Nature Methods (2024).

Case 4: SPACE-seq — A Unified Route to Spatial Chromatin and RNA

Problem: Lower barriers to combined profiling by leveraging standard whole-transcriptome capture reagents.

Data Setup: Spatial epigenomic libraries designed to be compatible with standard spatial transcriptomics capture.

Core Idea: Use polyA-tailed epigenomic libraries to enable spatial epigenomics and multiomics in a unified workflow.

Outputs: Spatial chromatin signals aligned to spatial RNA; joint clustering for region characterization.

Sanity Checks: Coverage sufficiency by region; expected concordance between accessibility and expression for known markers.

Key Reference: Huang et al. "Unified Molecular Approach for Spatial Epigenome, Transcriptome, and Cell Lineages." PNAS (2025).

Case 5: bindSC — Unbiased Integration as a Reference Framework

Problem: Avoid biased alignment that overfits shared features between modalities.

Data Setup: Single-cell multi-omics matrices with partial overlap.

Core Idea: Jointly align cells and features to reduce systematic integration bias.

Outputs: Cross-modality matching; integrated structure for baseline comparison.

Sanity Checks: Stability under feature selection changes; agreement with known biology.

Key Reference: Dou et al. "Unbiased Integration of Single Cell Multi-Omics Data." bioRxiv (2020).

Case 6: SIMO — Probabilistic Alignment for Multi-Omics Spatial Mapping

Problem: Place multi-omics single-cell states into tissue context using a spatial reference.

Data Setup: Multi-omics single-cell data plus spatial transcriptomics reference.

Core Idea: Estimate probabilistic mappings that balance similarity and spatial structure.

Outputs: Cell-to-space probability maps; spatial distribution of programs and states.

Sanity Checks: Confidence stratification; region-level plausibility; replicate stability.

Key Reference: Yang et al. "Spatial Integration of Multi-Omics Single-Cell Data with SIMO." Nature Communications (2025).

What to Report: Figures, Tables, and Reproducibility Notes

A strong report shows the biology and makes it clear why the alignment is trustworthy. It also separates measured layers from inferred layers, so readers can evaluate claims without guessing.

Deliverables Table

Figure 5. A practical deliverables checklist for reporting spatial multi-omics results with clarity.

Red Flags Worth Calling Out

• Domains track slide ID, processing day, or sequencing batch.
• Predicted layers look uniformly cleaner than measured layers.
• Rare niches disappear after filtering without justification.
• Small parameter changes flip domain identities.
• Chromatin and RNA signals contradict each other systematically without a plausible explanation.

A Short Project Brief Template (Helps Analysis Start Faster)

Projects move faster when the brief includes:

• One-sentence study goal (domains, enhancer prioritization, mapping).
• Modalities available and whether they are paired or unpaired.
• Tissue type, section count, and replicate plan.
• Preferred unit of analysis (spot, pixel, or segmented cell).
• Expected outputs (domains, enhancer–gene table, TF programs, figures).
• Key constraints (sample limits, compute environment requirements).

If you want your final report to support a mechanism-focused story (not just a domain map), share a brief summary of your study goal, whether your data are paired or unpaired, and the outputs you need (domains, enhancer–gene hypotheses, TF programs). CD Genomics can return a scoped proposal with transparent deliverables and pricing tiers aligned to your dataset size.

For high-resolution spatial RNA anchoring, you can also reference our 10 µm Spatial Transcriptomics Sequencing Service.

FAQs

How Do You Integrate Spatial Transcriptomics With Spatial ATAC-seq in Unpaired Studies?

Start with strict modality-specific QC, then use coordinate-based neighborhood graphs as a backbone for integration. Use an unpaired-capable method and report stability checks across parameter settings. Avoid presenting imputed layers as direct measurements.

What Validation Steps Make Unpaired Spatial Multi-Omics Integration Credible?

Show replicate concordance, demonstrate that domains are not explained by batch variables, and verify that known marker patterns are consistent with tissue anatomy. Include at least one sensitivity check figure that shows how results change under parameter shifts.

How Do You Build Enhancer–Gene Links Without Overclaiming Causality?

Treat enhancer–gene links as ranked hypotheses. Combine proximity constraints, co-variation with spatial expression, and motif context as supporting evidence, then prioritize a smaller set for follow-up experiments. Report what was measured versus inferred.

When Should You Choose Spatial CUT&Tag Instead of Spatial ATAC-seq?

Choose spatial CUT&Tag when the study hinges on specific histone marks or targeted protein–DNA contexts, and choose spatial ATAC-seq when broad accessibility landscapes are the main need. Match the assay to the regulatory question you want to answer before selecting an integration tool.

What Should Be Included in a Final Report for a Regulatory Mechanism Story?

Include joint domains, modality-specific markers, a ranked enhancer–gene table tied to regions, and a reproducibility package with versions, parameters, and QC summaries. Clearly label any predicted layers and provide uncertainty or confidence summaries.

Reference

1. Huang, Y. H., et al. "Unified Molecular Approach for Spatial Epigenome, Transcriptome, and Cell Lineages." Proceedings of the National Academy of Sciences, 2025, doi:10.1073/pnas.2424070122.
2. Li, Z., et al. "Integrative Deep Learning of Spatial Multi-Omics with SWITCH." Nature Computational Science, 2025, doi:10.1038/s43588-025-00891-w.
3. Long, Y., et al. "Deciphering Spatial Domains from Spatial Multi-Omics with SpatialGlue." Nature Methods, 2024, doi:10.1038/s41592-024-02316-4.
4. Hong, D., et al. "Divergent Combinations of Enhancers Encode Spatial Gene Expression." Nature Communications, 2025, doi:10.1038/s41467-025-60482-1.
5. Yang, P., et al. "Spatial Integration of Multi-Omics Single-Cell Data with SIMO." Nature Communications, 2025, doi:10.1038/s41467-025-56523-4.
6. Dou, J., et al. "Unbiased Integration of Single Cell Multi-Omics Data." bioRxiv, 2020, doi:10.1101/2020.12.11.422014.