Single Cell Deconvolution

1. Prerequisites

None.

1. Introduction

Multiple algorithms for single cell deconvolution have been published that allow to map single cell sequencing data sets in the spatial dimensions of a Visium data set generated from the same tissue. SPATA2 offers functions to visualize that and to integrate the positioning of cells in gradient screening. The data set used is an injured mouse cortex slice as has been published by Koupourtidou et al. 2023.

The injuries have been integrated in our analysis using the image annotations system. They were called inj1 and inj2.

library(SPATA2)
library(tidyverse)

# load data set 
object_mci <- downloadloadPubExample("MCI_LMU")

# set default and image annotations
object_mci <- setDefault(object_mci, clrp = "sifre", pt_clrp = "sifre")

data("image_annotations")

object_mci <- 
  setImageAnnotations(
    object = object_mci,
    img_anns = image_annotations[["MCI_LMU"]], 
    overwrite = TRUE
    )

injury_add_on <- 
  ggpLayerImgAnnOutline(object_mci, ids = c("inj1", "inj2"))

# get single cell data set 
data("sc_deconvolution")
sc_mci <- sc_deconvolution[["MCI_LMU"]]

# plot example data
plotSurface(object_mci, color_by = "clusters") + 
  injury_add_on

plotSurface(sc_mci, color_by = "cell_type", pt_clrp = "sifre", pt_size = 1.5) + 
  injury_add_on
Fig.1 The injured mouse cortex sample clustered and with deconvolution single cell data.Fig.1 The injured mouse cortex sample clustered and with deconvolution single cell data.

Fig.1 The injured mouse cortex sample clustered and with deconvolution single cell data.

2. Single Cell input

Most function that integrate single cell deconvolution take the single cell input via the argument sc_input. It expects a data.frame that contains at least the columns x and y as numeric pixel coordinates and a factor column called cell_type.

sc_mci
## # A tibble: 4,356 x 7
##        x     y cell_type             spot_id              barcodes   x_mm y_mm
##    <dbl> <dbl> <fct>                 <chr>                <chr>      [mm] [mm]
##  1  442.  378. Neurons               AAACAAGTATCTCCCA-1_0 AAACAAGTA~ 5.83 4.98
##  2  412.  140. Mural cells           AAACAGAGCGACTCCT-1_0 AAACAGAGC~ 5.42 1.85
##  3  412.  137. Astrocytes            AAACAGAGCGACTCCT-1_1 AAACAGAGC~ 5.43 1.81
##  4  413.  142. Astrocytes            AAACAGAGCGACTCCT-1_2 AAACAGAGC~ 5.44 1.87
##  5  412.  138. Microglia             AAACAGAGCGACTCCT-1_3 AAACAGAGC~ 5.43 1.82
##  6  424.  447. Monocytes/Macrophages AAACATTTCCCGGATT-1_0 AAACATTTC~ 5.58 5.89
##  7  425.  452. Microglia             AAACATTTCCCGGATT-1_1 AAACATTTC~ 5.60 5.96
##  8  426.  450. Microglia             AAACATTTCCCGGATT-1_2 AAACATTTC~ 5.62 5.93
##  9  492.  341. OPCs                  AAACCCGAACGAAATC-1_0 AAACCCGAA~ 6.48 4.50
## 10  494.  346. OPCs                  AAACCCGAACGAAATC-1_1 AAACCCGAA~ 6.51 4.56
## # i 4,346 more rows

Further columns are optional. For this vignette the three mentioned above suffice.

cell_types <- c("Astrocytes", "Monocytes/Macrophages", "Neurons", "Microglia")

sc_mci <- 
  filter(sc_mci, cell_type %in% {{cell_types}}) %>% 
  mutate(cell_type = droplevels(cell_type)) %>% 
  select(x, y, cell_type)

sc_mci
## # A tibble: 3,221 x 3
##        x     y cell_type            
##    <dbl> <dbl> <fct>                
##  1  442.  378. Neurons              
##  2  412.  137. Astrocytes           
##  3  413.  142. Astrocytes           
##  4  412.  138. Microglia            
##  5  424.  447. Monocytes/Macrophages
##  6  425.  452. Microglia            
##  7  426.  450. Microglia            
##  8  215.  391. Monocytes/Macrophages
##  9  215.  390. Astrocytes           
## 10  215.  393. Astrocytes           
## # i 3,211 more rows

3. Visualization

To visualize the cells in space use plotSurfaceSC().

injury_add_on_thin <- 
  ggpLayerImgAnnOutline(object_mci, ids = c("inj1", "inj2"), line_size = 0.5)

plotSurfaceSC(
  object = object_mci, 
  sc_input = sc_mci, 
  cell_types = cell_types, 
  display_image = TRUE, 
  display_density = FALSE,
  display_facets = TRUE, 
  pt_size = 0.5, 
) + 
  injury_add_on_thin + 
  legendNone()
Fig.2 Single cells of different cell types plotted on histology.

Fig.2 Single cells of different cell types plotted on histology.

2d density plots might be more insightful.

plotSurfaceSC(
  object = object_mci, 
  sc_input = sc_mci, 
  cell_types = cell_types, 
  display_image = TRUE, 
  display_density = TRUE,
  display_points = FALSE,
  display_facets = TRUE, 
  line_size = 0.5
) + 
  injury_add_on_thin + 
  legendNone()
Fig.3 2D kernel density estimates of single cells on the histology.

Fig.3 2D kernel density estimates of single cells on the histology.

The density of both, microglia and monocytes, seems to decrease rapidly with the distance to the image annotations. The contrary seems to apply to neurons. The density of astrocytes seems not to be significantly affected.

4. Spatial relation to image annotations

The density of specific cells might stand in relation to the distance to specific areas in the tissue. Figure 3 indicates high density of microglia and monocytes around the injury zones. As we have shown in our vignettes about gene expression and image annotations one can plot the abundance of X against the distance to image annotations, thus specific histological microstructures.

plotImageGgplot(object = object_mci) + 
  ggpLayerEncirclingIAS(
    object = object_mci, 
    id = c("inj1", "inj2"), 
    distance = "1mm"
  ) + 
  ggpLayerScaleBarSI(
    object = object_mci, 
    sb_dist = "1mm", 
    sb_pos = c("1.5mm", "6.75mm"), 
    sgmt_size = 0.75, 
    text_size = 7.5
  )
Fig.4 Visualization of the conducted screening of the area around the injury annotations.

Fig.4 Visualization of the conducted screening of the area around the injury annotations.

Let X be the density of cells.

plotIasRidgeplotSC(
  object = object_mci, 
  id = c("inj1", "inj2"), # infer for both and then take the average
  distance = "1mm",
  sc_input = sc_mci, 
  free_y = TRUE,
  display_border = TRUE,
  include_area = TRUE # include the area of the injury annotations
) + 
  labs(subtitle = "a) Free y-axis") + 
  legendNone()

plotIasRidgeplotSC(
  object = object_mci, 
  id = c("inj1", "inj2"), 
  distance = "1mm",
  sc_input = sc_mci,
  free_y = FALSE, 
  display_border = TRUE,
  include_area = TRUE 
) + 
  labs(subtitle = "b) Fixed y-axis") + 
  legendNone()
Fig.4 Inferred density of cell types as a function of distance to the injury annotations.Fig.4 Inferred density of cell types as a function of distance to the injury annotations.

Fig.4 Inferred density of cell types as a function of distance to the injury annotations.

Use the function inferSingleCellGradient() to obtain the results in a data.frame.

inferSingleCellGradient(
  object = object_mci, 
  sc_input = sc_mci, 
  id = c("inj1", "inj2"), 
  distance = "1mm",
  calculate = "density", # compute density 
  area_unit = "mm2", # with unit count/mm2
  normalize = FALSE # do not rescale to 0-1
)
## # A tibble: 11 x 7
##    bins_circle bins_order bins_angle Neurons Astrocytes Microglia
##    <fct>            <dbl> <fct>        <dbl>      <dbl>     <dbl>
##  1 Core                 0 (0,360]        0        114.      240. 
##  2 Circle 1             1 (0,360]       42.1       87.9     202. 
##  3 Circle 2             2 (0,360]       26.9       83.7     105. 
##  4 Circle 3             3 (0,360]       47.2       97.4      57.4
##  5 Circle 4             4 (0,360]       48.1      108.       73.7
##  6 Circle 5             5 (0,360]       35.3       84.2      74.3
##  7 Circle 6             6 (0,360]       52.1      108.       47.3
##  8 Circle 7             7 (0,360]       58.2      119.       45.5
##  9 Circle 8             8 (0,360]       49.8       87.0      67.8
## 10 Circle 9             9 (0,360]       56.4       79.5      40.8
## 11 Circle 10           10 (0,360]       42.0       64.4      52.4
## # i 1 more variable: `Monocytes/Macrophages` <dbl>

The columns named by the cell type contain the density values calculated for each distance/circular bin in the unit specified via area_unit. Use the argument as_models = TRUE to obtain the results as valid input for add_models in image annotation screening.

inferSingleCellGradient(
  object = object_mci, 
  sc_input = sc_mci, 
  id = c("inj1", "inj2"), 
  distance = "1mm",
  calculate = "density",
  as_models = TRUE
)
## $Neurons
##  [1] 0.0000000 0.7229662 0.4624053 0.8106138 0.8261313 0.6065198 0.8960836
##  [8] 1.0000000 0.8564700 0.9695214 0.7225793
## 
## $Astrocytes
##  [1] 0.9022810 0.4304072 0.3522829 0.6027643 0.7881667 0.3631900 0.7889645
##  [8] 1.0000000 0.4141949 0.2756438 0.0000000
## 
## $Microglia
##  [1] 1.00000000 0.80944945 0.32133099 0.08314482 0.16503405 0.16838363
##  [7] 0.03270634 0.02345116 0.13558303 0.00000000 0.05814409
## 
## $`Monocytes/Macrophages`
##  [1] 1.00000000 0.45156204 0.11913118 0.09565253 0.11272233 0.08934293
##  [7] 0.05290540 0.03235082 0.00000000 0.07798454 0.03589546

5. Accounting for multiple sections

Note that this sample contains two sections that, albeit being from the same organ, are independent of each other. Relating cells from one section to the image annotation of another section does not make any sense. The function add_tissue_section_variable() and include_tissue_outline() account for that as is exemplified in this visualization with a distance set to 6mm which would go beyond the capture frame of the Visium slide.

# visualizes the result of include_tissue_outline()
tissue_outline <- 
  ggpLayerTissueOutline(object_mci, line_color = "black", line_size = 1)

# uses include_tissue_outline()
many_expansions <- 
  ggpLayerEncirclingIAS(
    object = object_mci,
    id = c("inj1", "inj2"), 
    distance = "6mm", 
    line_size = 1
    )

ggplot() +
  ggpLayerImage(object_mci) + 
  theme_bw() + 
  coord_equal() +
  tissue_outline + 
  ggpLayerThemeCoords()

ggplot() + 
  ggpLayerImage(object_mci) + 
  theme_bw() + 
  coord_equal() +
  tissue_outline +
  many_expansions + 
  ggpLayerThemeCoords()
Fig.6 Visualization of how SPATA2 considers the outline of actual tissue sections within algorithms.Fig.6 Visualization of how SPATA2 considers the outline of actual tissue sections within algorithms.

Fig.6 Visualization of how SPATA2 considers the outline of actual tissue sections within algorithms.