Spatial Transcriptomics Meets Single‑Cell RNA‑seq: Rebuilding Tissue Context Without Losing Resolution

By EVOBYTE: Your partner in bioinformatics

Introduction

Single‑cell RNA‑seq (scRNA‑seq) gives you a detailed inventory of cell states, but it tears cells out of their neighborhoods. Spatial transcriptomics (ST) preserves where those cells live, yet it often trades off sensitivity or whole‑transcriptome coverage. Put them together, and you can see who’s in the tissue and where they are—critical for questions in oncology, immunology, and pathology such as “Which T cells sit next to which tumor clones?” and “What signals flow across that boundary?” Spatial integration is quickly moving from a nice‑to‑have to a standard for target discovery and biomarker development.

Two Core Strategies: Deconvolution vs. Mapping

When teams say they “integrate spatial and single‑cell data,” they typically mean one of two things.

First, deconvolution. Many ST assays (for example, early Visium slides) capture transcripts from multiple cells per spot. Deconvolution uses an scRNA‑seq reference to estimate the proportions of cell types in each spot. Methods like RCTD and cell2location fall here; they’re great for quantifying cell‑type admixture across tissue regions and are widely used to profile tumor microenvironments or inflamed tissue.

Second, mapping. Here you assign single cells directly to spatial coordinates, effectively projecting scRNA‑seq annotations into tissue space. Tangram is a prominent example, aligning sc/snRNA‑seq to spatial data (Visium, MERFISH, and others) to reconstruct fine‑grained spatial maps. In benchmarking studies, mapping methods like Tangram perform strongly for transcript distribution prediction, while deconvolution tools such as cell2location and SpatialDWLS often lead for estimating cell‑type mixtures—so choose based on your question.

A quick note on technology trade‑offs: no single platform simultaneously delivers unlimited gene plex, high sensitivity, and true single‑cell spatial resolution across large tissue areas. This is precisely why computational integration with scRNA‑seq is so valuable.

A Practical, Reproducible Workflow (R and Python)

Let’s ground this with a minimal, end‑to‑end outline you can adapt.

R (Seurat v5 + RCTD) for spot‑level deconvolution using an scRNA‑seq reference and Visium data:

library(Seurat)
library(spacexr)   # RCTD
# 1) Load data
visium <- Load10X_Spatial("path/to/visium")      # ST counts + image
sc     <- readRDS("path/to/singlecell_reference.rds")
# 2) Preprocess
visium <- SCTransform(visium, verbose = FALSE)
sc     <- SCTransform(sc, verbose = FALSE)
# 3) Build RCTD objects and run
ref <- Reference(sc@assays$SCT@counts, sc$celltype)
sp  <- SpatialRNA(as.matrix(visium@assays$SCT@counts), visium@images[[1]]@coordinates)
rctd <- create.RCTD(sp, ref)
rctd <- run.RCTD(rctd, doublet_mode = "doublet")
deconv <- get_deconvolution_results(rctd)

This yields per‑spot cell‑type proportions you can overlay on the H&E image to see tissue architecture and gradients.

Python (Tangram) for cell‑to‑space mapping:

import tangram as tg
import scanpy as sc
# 1) Load single-cell and spatial objects
adata_sc = sc.read_h5ad("sc_ref.h5ad")     # with annotated cell types
adata_sp = sc.read_h5ad("visium_st.h5ad")  # with spatial coordinates
# 2) Select marker genes and align
tg.pp_adatas(adata_sc, adata_sp, genes=None)
ad_map = tg.map_cells_to_space(adata_sc, adata_sp, mode="cells")
# 3) Project cell labels into space
tg.project_cell_annotations(ad_map, adata_sp, annotation="cell_type")

You’ll get spatial maps of predicted cell labels and expression, useful for neighborhood analysis and hypothesis generation.

Quality Tips and Industry Relevance

Success depends on three practical choices.

Data compatibility. Match chemistry and biology between datasets as closely as possible. FFPE vs. fresh‑frozen, targeted imaging panels (e.g., MERFISH, Xenium) versus sequencing‑based whole‑transcriptome assays (e.g., Visium/Visium HD) can influence which genes overlap and how well signals transfer. Plan for that at experimental design time, not after sequencing.

Model selection tied to the hypothesis. If you need precise cell‑type proportions in regions (say, to quantify T‑cell infiltration across tumor margins), deconvolution is often the right tool. If you want to reconstruct spatial trajectories or discover niches (e.g., stromal–immune hubs), mapping approaches shine. Blending both often clarifies the picture, especially in complex tissues.

Validation matters. Always sanity‑check spatial predictions with orthogonal evidence: known histology, spatially variable genes, or targeted in situ panels. Contemporary vignettes in Seurat show how to layer these checks and handle large data efficiently, which is essential for production‑grade analysis in pharma and diagnostics pipelines.

Summary / Takeaways

Key idea: combine scRNA‑seq depth with ST context to see not just which cells are present, but where they act.
Two paths: deconvolution estimates cell‑type mixtures per spot; mapping projects single cells into space.
Start simple: use Seurat v5 + RCTD for deconvolution or Tangram for mapping, then validate with histology or targeted panels.
Business impact: sharper patient stratification, spatial biomarkers, and better target localization—faster decisions with higher confidence.