Skip to contents

Quick start

Source: vignettes/Quick_start.Rmd
Quick_start.Rmd
library(SpaTopic)
library(ggplot2)
my_colors <- c("#66b803", "#E5AA9B", "#FABD3E", "#2B8CBE", "#DE77AE", "#9970AB", "gray", "#D5E4A2", "#71D0F5", "#B1746F", "#ADE2D0", "#20DE8BFF", "#CCDE8BFF", "#FFDE8BFF", "#FFA88BFF", "#FF6A8BFF")

Load data

SpaTopic requires two inputs:

  • spot_celltype: The result of the deconvolution of single-cell and spatial transcriptome data, or a matrix with rows as spots and columns as cell types.*

  • spot_clusters: A dataframe with spot domain information.*

And they should be like

data("spot_celltype")
data("spot_clusters")
head(spot_celltype)
      Acinar_cells Ductal_cells Cancer_clone_A Cancer_clone_B         DCs
10x10 5.838572e-02    0.2349066   1.365076e-03   3.892868e-04 0.165860789
10x13 4.807943e-05    0.9984677   1.654640e-06   9.032885e-06 0.001244634
10x14 4.701190e-02    0.8373601   4.846860e-03   9.009235e-04 0.003541947
10x15 5.047613e-02    0.8020465   1.911570e-04   3.325224e-02 0.084113110
10x16 4.694120e-03    0.9718078   1.719378e-06   6.266388e-04 0.007665514
10x17 6.850619e-02    0.7737652   1.034929e-02   5.242836e-02 0.030090839
        Tuft_cells         pDCs Endocrine_cells Endothelial_cells  Macrophages
10x10 1.149557e-02 5.750904e-05    3.794494e-03      1.554314e-01 7.575614e-02
10x13 1.411944e-04 2.179189e-08    2.791803e-08      2.522234e-08 4.952721e-05
10x14 2.430843e-05 6.745099e-05    7.014669e-08      9.208190e-03 3.539390e-02
10x15 3.087975e-07 2.387285e-04    1.660241e-07      3.787078e-04 1.868701e-02
10x16 2.859347e-07 1.835252e-04    9.321908e-09      1.535896e-04 2.258778e-05
10x17 7.639672e-04 1.963630e-04    1.888457e-04      2.610257e-03 1.254141e-03
        Mast_cells T_cells_._NK_cells    Monocytes         RBCs  Fibroblasts
10x10 1.335059e-07       1.419896e-05 6.324585e-05 6.935704e-08 2.924798e-01
10x13 2.560977e-08       8.100230e-10 2.999011e-05 4.812122e-06 3.255235e-06
10x14 7.319528e-06       2.780869e-06 8.047062e-04 7.357054e-06 6.082224e-02
10x15 9.115675e-07       2.629728e-06 4.568320e-04 2.846073e-05 1.012715e-02
10x16 6.924793e-07       2.044994e-05 1.654387e-06 1.444002e-09 1.482138e-02
10x17 4.102771e-06       1.894866e-04 6.179414e-04 3.425166e-06 5.903159e-02
head(spot_clusters)
      row col sizeFactor cluster.init spatial.cluster
10x10  10  10  4.7761108            1               2
10x13  10  13  1.0052199            2               2
10x14  10  14  0.8106812            2               2
10x15  10  15  0.4987377            2               2
10x16  10  16  0.4346143            2               2
10x17  10  17  0.7446991            2               2

*The data used here is from CARD (spot_celltype) and BayesSpace (spot_clusters).

About data

The corresponding Seurat object data can also be obtained from GEO. (Data used here is processed by SCTransform) The spot_celltype and spot_clusters is based on it.

load("vignette_data/PDAC_obj.rda")
PDAC_obj <- AddMetaData(PDAC_obj, spot_clusters["spatial.cluster"])
DimPlot(PDAC_obj, group.by = 'spatial.cluster', reduction = "umap",cols = my_colors)
SpatialDimPlot(PDAC_obj, label = FALSE, group.by = "spatial.cluster", pt.size.factor = 8) + scale_fill_manual(values = my_colors)

Simply usage

Enter the deconvolution result spot_celltype and the grouping information of the spot spot_clusters to get an result_list :

  • CellTopic is a dataframe which can be add to a Seurat object.

  • domain_topic is a dataframe, row is CellTopics and col is domains.

  • celltype_topic is a dataframe, row is cell types and col is CellTopics.

  • Cell_topic is a vector of which topic can be chosen in each CellTopic.

For subsequent analyses, CellTopic and celltype_topic are mainly used.

result_list <- CellTopic(
  spot_celltype,
  spot_clusters,
  cluster = "spatial.cluster",
  num_topics = 13
)

Plot

You can add the results of CellTopic to meta.data of Seurat object through AddMetaData of Seurat, and display the spatial position of all CellTopics through SpatialDimPlot of Seurat.

PDAC_obj <- AddMetaData(PDAC_obj, result_list[["CellTopic"]])
SpatialDimPlot(PDAC_obj, group.by = "CellTopic",pt.size.factor = 8)

And CellTopic_plot can be used to display the spatial location and cell type of one or more CellTopics at the same time.

Please use the Seurat object with CellTopic added as meta.data and the celltype_topic dataframe in the result list of FindCellTopic (or CellTopic) function as input.

CellTopic_plot(PDAC_obj, 
               celltype_topic = result_list[["celltype_topic"]], 
               cols.highlight = c("#DE2D26", "#FEE0D2"),
               cols.celltype = my_colors, 
               pt.size.factor = 8)

Optional parameters 

result_list <- CellTopic(
  spot_celltype,
  spot_clusters,
  cluster = "spatial.cluster",
  num_topics = 13,
  percent = 0.7
)

We recommend that the value of the num_topics is generally slightly larger than the group number of spots.

The larger the percent, the more dependent the results of deconvolution. We recommend using a value of 0.5-0.9.

Another strategy

If you want to focus less on percentages and more on the information carried by a single topic, you can use Binarization = TRUE .

result_list <- CellTopic(
  spot_celltype,
  spot_clusters,
  cluster = "spatial.cluster",
  num_topics = 13,
  Binarization = TRUE
)
PDAC_obj <- AddMetaData(PDAC_obj, result_list[["CellTopic"]])
CellTopic_plot(PDAC_obj, 
               celltype_topic = result_list[["celltype_topic"]], 
               cols.highlight = c("#DE2D26", "#FEE0D2"),
               cols.celltype = my_colors, 
               pt.size.factor = 8)

Meta Topic

If you want to explore different spatial domains based on the cell type of the CellTopic, we provide a simple clustering option, meta.cell = TRUE.

result_list <- CellTopic(
  spot_celltype,
  spot_clusters,
  cluster = "spatial.cluster",
  num_topics = 13,
  percent = 0.7,
  meta.cell = TRUE,
  k = 3
)
head(result_list[["MetaTopic"]])
           MetaCluster
CellTopic1           1
CellTopic2           1
CellTopic3           2
CellTopic4           3

Parameter k is a integer number of MetaTopics. Hierarchical clustering algorithm is used by default