Skip to contents

Visualization

spacedeconv_visualization.Rmd

To introduce spacedeconvs visualization functions we will utilize a deconvolution result obtained from the sample data provided and the deconvolution method spatialdwls. First we compute a reference signature based on the available scRNA-seq data with cell-type annotation. Then we quantify cell-type fractions in the 10X Visium slide using the computed reference signature. 

library(spacedeconv)
data("spatial_data_3")
data("single_cell_data_3")

single_cell_data_3 <- spacedeconv::preprocess(single_cell_data_3)
spatial_data_3 <- spacedeconv::preprocess(spatial_data_3)

spatial_data_3 <- spacedeconv::normalize(spatial_data_3, method = "cpm")

signature <- spacedeconv::build_model(
  single_cell_obj = single_cell_data_3,
  cell_type_col = "celltype_major",
  method = "spatialdwls", verbose = T
)
## class of selected matrix:  dgCMatrix 
## [1] "finished runPCA_factominer, method == factominer"
deconv <- spacedeconv::deconvolute(
  spatial_obj = spatial_data_3,
  single_cell_obj = single_cell_data_3,
  cell_type_col = "celltype_major",
  method = "spatialdwls",
  signature = signature,
  assay_sp = "cpm"
)
## class of selected matrix:  dgCMatrix 
## [1] "finished runPCA_factominer, method == factominer"

Available Visualizations

  1. plot_spatial() generates a Hex Plot of a SpatialExperiment containing deconvolution results.
  2. plot_gene() plot spatial gene expression.
  3. plot_umi_count() Plots the number of sequenced reads per spot.
  4. plot_most_abundant() Render a plot containing the most abundant cell-type for each spot.
  5. plot_comparison() Plot comparison of two cell-type fractions.

1. plot_spatial()

Plot any spot annotation with a continuous or discrete scale. The spot annotation needs to be of colData(spe), so manual annotation can be added to the SpatialExperiment object for visualization. 

spacedeconv::plot_spatial(deconv, result = "spatialdwls_B.cells", density = F, smooth = T, title="B cells")

2. plot_gene()

Plot gene expression on a spatial level. It may make sense to smooth the plot to simplify the detection of expression patterns. You can further select the assay using the assay parameter. 

spacedeconv::plot_gene(deconv, gene = "CD19", density = F, smooth = T, title_size=12)

3. plot_umi_count()

This plot shows the number of detected UMIs for each spot. We recommend rendering this plot with transform_scale = "sqrt" due to the large range of UMI count values. 

spacedeconv::plot_umi_count(deconv, transform_scale = "sqrt", density = F, smooth = T, title_size=12)

4. plot_most_abundant()

This plots displays the most abundant cell-type for each spot. You can specify which cells to plot by either one of the following:

  • cell_type vector of celltypes to plot
  • method plotting all cell types of the provided method
  • remove vector of celltypes to be removed from the plot 
spacedeconv::plot_most_abundant(deconv,
  method = "spatialdwls",
  density = F,
  title_size = 12,
  legend_size = 15,
  font_size = 10,
  remove = c("spatialdwls_Cancer.Epithelial", "spatialdwls_CAFs")
)

5. plot_comparison()

Use this function to plot the ratio of deconvolution results from two cell-types. 

spacedeconv::plot_comparison(deconv,
  cell_type_1 = "spatialdwls_Cancer.Epithelial", # red
  cell_type_2 = "spatialdwls_Normal.Epithelial", # blue
  palette = "Blue-Red",
  density = F,
  smooth = T,
  title_size=12
)

Color Palette

All palettes from the colorspace R Package can be used. 

colorspace::hcl_palettes(plot = TRUE)

Further plot adjustments

Image Alignment Offset

spacedeconvs Visualization function is designed to work with data by SpaceRanger >= V2.0. Since this Version the image is rotated by default that the hourglass fiducial is in the upper left corner. Previous SpaceRanger results can be rotated differently. The rotation additionally reflects in the angle of the spots on the slide. Uncorrectly rotated images result in hexagons being rotated by 30 degrees. To compensate for this you can use the offset_rotation parameter to correct the hexagon alignment. This is only necessary for Visium slides where the hourglass fiducial is in the bottom left or upper right corner. 

plot_umi_count(deconv, offset_rotation = T, title_size=12) # rotate hexagons

Transform Scale

With the transform_scale parameter the colorspace scale can be modified. Available options are: “ln”, “log10”, “log2” and “sqrt”. Scaling the color range differently can aid with interpreting the plot. Please have in mind that the plot does not show valid deconvolution results anymore and should be handled with caution. 

spacedeconv::plot_umi_count(deconv, transform_scale = "sqrt", density = FALSE, title_size=12)

Smooth

With this parameter the expression values can be smoothed to simplify pattern recognition. The smoother utilizes a linear kernel which size is calculated by multiplying the spot distance with the smoothing factor. It has to be mentioned that this operation changes the deconvolution result in the final plots by appling the kernel, so they should be interpreted as a helpful visualization option and not a deconvolution result.

  • smooth=T enable smoothing
  • smoothing_factor choose kernel size (factor of spot distance) 
spacedeconv::plot_spatial(deconv,
  result = "spatialdwls_B.cells",
  smooth = T,
  smoothing_factor = 1.5,
  density = FALSE,
  title="B cells smoothed"
)

Density Distribution

You can add a density distribution by setting density = TRUE. The red line corresponds to the mean of the distribution. 

spacedeconv::plot_spatial(deconv,
  result = "spatialdwls_B.cells",
  smooth = TRUE,
  density = TRUE,
  title="B cells with density"
)

Save Plots

You can save a plot by setting save=TRUE. Specify a file path with the path parameter. If no path is provided your plots will be stored at ~/spacedeconvResults/. To change the size of the rendered plot use png_width and png_height to set the plot size in pixels. Plots are saved as a png. 

spacedeconv::plot_spatial(deconv,
  result = "quantiseq_B.cell",
  smooth = TRUE,
  density = TRUE,
  save = TRUE,
  path = "~/project/results",
  png_width = 1000,
  png_height = 750
)

Aggregate cell types

Aggregate cell types using the aggregate function. Internall the deconvolution results are summed up and get a new name. The aggregation can be visualized with all available plotting functions. 

spe <- aggregate(spe, cell_type_1, cell_type_2, newName)

Additional Parameters

  • show_image logical, show or remove the spatial image
  • spot_size integer, increase (>1) or decrease (<1) the hexagon size
  • limits vector containing upper and lower limits for the color scale
  • palette_type “discrete”, “sequential” or “divergent”, how to scale the color´
  • reverse_palette reverse color palette
  • font_size font size of legend
  • title_size font size of title
  • legend_size legend
  • title set a custom title