Spatial Transcriptomics Analysis

1. Intro to Spatial Omics

Traditional single-cell RNA sequencing requires tissues to be dissociated, leading to a complete loss of spatial context. Spatial transcriptomics solves this by capturing both the gene expression profile and the physical location of cells within a tissue section, allowing us to map biology directly onto histology.

2. Key Technologies

There are broadly two categories of spatial technologies, each requiring different analytical approaches:

• Sequencing-based (e.g., Visium): Uses barcoded spots on a slide. Typically lower resolution (spots contain 1-10 cells) but captures the whole transcriptome unbiasedly.
• Imaging-based (e.g., Xenium, MERFISH, CosMx): Uses in situ hybridization and fluorescence. Offers true single-cell or sub-cellular resolution, but usually targeted to a specific panel of genes (e.g., 500-5000 genes).

Data representation uses frameworks like `SpatialData` (Python) or `SpatialExperiment` (R) to handle the complexities of images, coordinates, and count matrices simultaneously.

3. Quality Control

QC in spatial transcriptomics involves both standard count-based metrics and image-based checks:

• Filtering out spots/cells with low total counts or few detected genes.
• Removing spots that fall outside the actual tissue boundary on the image.
• Checking for spatial artifacts or edge effects introduced during tissue sectioning or permeabilization.

4. Spatial Domains

Similar to clustering in scRNA-seq, we want to identify distinct regions within the tissue (e.g., tumor core, invasive margin, healthy stroma).

However, advanced spatial clustering algorithms (like SpaGCN or Seurat's spatial features) incorporate both transcriptional similarity and spatial proximity, ensuring that the identified domains form coherent physical regions on the slide.

5. Cell Type Deconvolution

For technologies where spots capture multiple cells (like 10x Visium), we cannot directly assign a single cell type to a spot. Instead, we use Deconvolution algorithms.

These tools (e.g., Cell2location, RCTD) use an annotated single-cell reference atlas to estimate the proportion or absolute abundance of different cell types within each spatial spot. This maps high-resolution cell type definitions onto the lower-resolution spatial map.

6. Neighborhood Analysis

Once we have single-cell resolution (either natively or via deconvolution), we can ask complex biological questions about the microenvironment: