Spatial extension

The scdrake package now provides spatial extensions for both the first stage (01_input_qc) and the second stage (02_norm_clustering) of the single-sample pipeline.
These spatial workflows are currently optimized for Visium data and other spot-based spatial transcriptomics technologies.

The aim is to offer functionality comparable to Seurat, Giotto (R), and scanpy (Python), while following the principles of the Orchestrating Spatial Transcriptomics Analysis with Bioconductor guide.

For advanced analysis of spatial datasets—particularly for cell–cell interaction and communication region identification we recommend: - CellChat2 (R) - LIANA+ (Python)

The latter is accompanied by the vignette vignette("interaction_analysis"), which provides a step-by-step example of exporting count matrices and metadata from SpatialExperiment objects in R, and then importing them into Python to create an AnnData object.

This vignette is a supplement to: - vignette("stage_input_qc") - vignette("stage_norm_clustering")

Spatial extention functions

Input data

If the spaceranger option is selected in the the first stage (01_input_qc) for reading input data, the data are loaded as SpatialExperiment.

Spatial visualization

For (01_input_qc) and (02_norm_clustering) of the single-sample pipeline we now offer visualization of tissue, as pseudo tissue spot visualization from ggspavis. For (01_input_qc) stage, quality control metrics can be overlaid directly onto the H&E-stained tissue image, enabling visual inspection of tissue quality and spatial artifacts. In other section only spots are visualized for greater clarity. For downstream sections, visualization defaults to pseudo-tissue spot plots (using ggspavis), where spots are displayed without the underlying image for greater clarity.

Selection of spatially variable genes

If the SVG option is enabled in stage 02_norm_clustering (see vignette("stage_norm_clustering")), spatially variable genes are identified alongside highly variable genes (HVGs). That is done using nnSVG::nvSVG, with the selection method MoransI. A straightforward union of SVGs and HVGs is taking to further processing, see https://www.biorxiv.org/content/10.1101/2021.08.27.457741v1 for more details.

Marker-based annotation

Marker-based annotation was implemented for both single-cell and spatial datasets. In summary, expression profiles and statistical metrics are computed for each cell/spot, the result is visualized using a heatmap and dimension reduction plot. For spatial datasets is enabled to visualized results in tissue coordinates, both enrichment plots for each annotation label (individual enrichment plots) and for overall results for each spot. This functionality is adapted from the Giotto package, with methods based on Kim et al., 2005.

Spot deconvolution

CARD and RCTD can be use for spot deconvolution. CARD uses a Bayesian hierarchical model that incorporates spatial correlation between spots, while RCTD applies a generalized linear model for a more direct, assumption-light fit. We include both methods to give users the choice between a spatially smoothed approach (CARD) and a faster, model-driven method (RCTD), depending on dataset characteristics and analysis goals.