This R package contains the code to perform the SPatial Omnibus Test (SPOT). SPOT takes a range of radii, starting with 0, calculates a spatial summary of the point locations across images, and calculates an omnibus p-value describing the strength of association between the spatial summary and the outcome across radii.
For more information, see our publication in Bioinformatics: https://academic.oup.com/bioinformatics/article/40/7/btae425/7702328
To install SPOT, run the following:
devtools::install_github("sarahsamorodnitsky/SPOT")
library(SPOT)
This package requires the following dependencies: spatstat, survival, ACAT, dplyr, tidyselect, svMisc.
We will illustrate use of SPOT on an ovarian cancer dataset, which we download using the following code.
load(url("http://juliawrobel.com/MI_tutorial/Data/ovarian.RDA"))
This will load an object called ovarian_df into the R environment.
We first give the cells phenotypes based on their marker positivity. We also add a PID column based on the image IDs (this dataset contained only one image per person) and filter to only the cells detected within tumors (rather than the stroma region).
# Label the immune cell types
ovarian_df <- ovarian_df %>%
mutate(type = case_when(
phenotype_cd68 == "CD68+" ~ "macrophage",
phenotype_cd19 == "CD19+" ~ "B cell",
phenotype_cd3 == "CD3+" & phenotype_cd8 == "CD8-" ~ "CD4 T cell",
phenotype_cd3 == "CD3+" & phenotype_cd8 == "CD8+" ~ "CD8 T cell",
phenotype_ck == "CK+" ~ "Tumor"
))
# Remove any cell types with NA labels (corresponding to other cells)
ovarian_df <- ovarian_df %>% filter(!is.na(type))
# Change colnames
# According to VectraPolarisData vignette, sample_id is the image identifier, also subject id for the ovarian data
# Link: https://bioconductor.org/packages/devel/data/experiment/vignettes/VectraPolarisData/inst/doc/VectraPolarisData.html#52_HumanOvarianCancerVP
colnames(ovarian_df)[colnames(ovarian_df) == "sample_id"] <- "id"
# Add a PID column
ovarian_df$PID <- ovarian_df$id
# Filter to just cells detected within the tumor
ovarian_df_tumor <- ovarian_df %>% filter(tissue_category == "Tumor")
We then construct a series of radii at which to evaluate the spatial summary using Ripley's rule of thumb.
# Check the dimensions of each image
smallest.dim <- ovarian_df_tumor %>%
dplyr::group_by(id) %>%
dplyr::summarize(x.range = abs(max(x) - min(x)),
y.range = abs(max(y) - min(y)),
min = min(x.range, y.range))
radii.ripleys.rule <- seq(0, 0.25 * min(smallest.dim$min), length.out = 100)
We will examine the colocalization of immune cell types, including CD4 T cells, macrophage, CD8 T cells, and B cells. We will iterate through each combination of immune cell and assess the strength of the relationship between cell colocalization and survival.
First, we save the immune cell types and construct a list of all pairwise combinations.
# Create all combinations of immune cell types
cell.types <- unique(ovarian_df_tumor %>% filter(immune == "immune") %>% select(type) %>% unlist())
cell.type.combos <- combn(cell.types, 2, simplify = FALSE)
We then initialize a matrix, colocalization.matrix.tumor to store the SPOT p-value for the association between cell type colocalization and survival. We also create a list to store the output from SPOT.
# Initialize a matrix to store the results
colocalization.matrix.tumor <-
matrix(nrow = length(cell.types), ncol = length(cell.types),
dimnames = list(cell.types, cell.types))
# Save the metadata
colocalization.metadata.tumor <-
matrix(list(), nrow = length(cell.types), ncol = length(cell.types),
dimnames = list(cell.types, cell.types))
# Iterate through the cell type combinations
for (i in 1:length(cell.type.combos)) {
svMisc::progress(i/(length(cell.type.combos)/100))
# Save the combination
cell.type.i <- cell.type.combos[[i]]
# Save the row and column index
row.ind <- which(cell.types == cell.type.i[1])
col.ind <- which(cell.types == cell.type.i[2])
# Run SPOT
res <- spot(data = ovarian_df_tumor %>% filter(immune == "immune"),
radii = radii.ripleys.rule,
outcome = "survival_time", censor = "death",
use.K = FALSE, K.diff = FALSE, adjustments = "age_at_diagnosis",
model.type = "survival", cell.type = cell.type.i,
marked = TRUE, pick.roi = "all", print.progress = FALSE)
# Save the result
colocalization.matrix.tumor[row.ind, col.ind] <- res$overall.pval
colocalization.metadata.tumor[row.ind, col.ind][[1]] <- res
}
# Colocalization with tumor cells
tumor.colocalization.pvals <- gdata::unmatrix(colocalization.matrix.tumor)
tumor.colocalization.pvals <- tumor.colocalization.pvals[!is.na(tumor.colocalization.pvals)]
sort(p.adjust(tumor.colocalization.pvals, method = "fdr"))
# Create a table with the results
colocalization.matrix.tumor.df <- reshape2::melt(colocalization.matrix.tumor, na.rm = TRUE)
colnames(colocalization.matrix.tumor.df) <- c("Cell Type 1", "Cell Type 2", "SPOT P-Value")
colocalization.matrix.tumor.df$`SPOT Q-Value` <- p.adjust(colocalization.matrix.tumor.df$`SPOT P-Value`, method = "fdr")
# View the results:
colocalization.matrix.tumor.df[order(colocalization.matrix.tumor.df$`SPOT Q-Value`, decreasing = FALSE),]
# Cell Type 1 Cell Type 2 SPOT P-Value SPOT Q-Value
# CD4 T cell macrophage 0.007143277 0.02858606
# macrophage B cell 0.009528685 0.02858606
# CD4 T cell CD8 T cell 0.188707711 0.37741542
# macrophage CD8 T cell 0.395898161 0.59384724
# CD4 T cell B cell 0.921636024 0.92163602
# CD8 T cell B cell 0.919063591 0.92163602
We can further visualize the results using the ggplot2 package. We focus on CD4 T cell and macrophage colocalization:
colocalization.metadata.tumor[1,2][[1]]$pval.df %>%
mutate(pval.neg.log10 = -log10(pval)) %>%
ggplot(aes(x = radius, y = pval.neg.log10)) +
geom_point(alpha = 0.5) +
theme_bw() +
xlab("Radius") +
ylab("-log10(P-Value)") +
ggtitle("CD4 T Cells and Macrophages: P-Values for Association Between L(r) and \n Overall Survival at each Radius") +
geom_hline(yintercept = -log10(0.05), lty = 2)