CRAWDAD enables the characterization of spatial cell-type relationships across different length scales.
CRAWDAD is a statistical framework that uses cell-type labeled spatial omics data to identify the colocalization or separation of cell types at different length scales. CRAWDAD identifies the spatial relationship of cell types in the tissue and the length scale in which they reach significance. It identifies groups of cells with similar spatial relationships patterns. Also, CRAWDAD subsets the cell types based on their proximity to others. Lastly, CRAWDAD compares multiple tissue samples using their cell-type spatial relationship patterns. Therefore, CRAWDAD is a powerful tool for tissue characterization and comparison.
To install CRAWDAD, we recommend using remotes:
require(remotes)
remotes::install_github('JEFworks-Lab/CRAWDAD')library(crawdad)
library(tidyverse)
library(dplyr)
library(ggplot2)
## Load the spleen data of the pkhl sample
data('pkhl')
## Load the spleen data of the ksfb sample
data('ksfb')
all_celltypes <- unique(c(pkhl$celltype, ksfb$celltype))
colors <- rainbow(length(all_celltypes))
names(colors) <- all_celltypes
## Plot 1: pkhl dataset
p1 <- pkhl %>%
ggplot(aes(pkhl$x, pkhl$y, color = pkhl$celltype)) +
geom_point(size = .01) +
scale_color_manual(values = colors) +
guides(colour = guide_legend(override.aes = list(size = 1))) +
labs(title = 'pkhl') +
theme_minimal()
## Plot 2: ksfb dataset
p2 <- ksfb %>%
ggplot(aes(ksfb$x, ksfb$y, color = ksfb$celltype)) +
geom_point(size = .01) +
scale_color_manual(values = colors) +
guides(colour = guide_legend(override.aes = list(size = 1))) +
labs(title = 'ksfb') +
theme_minimal()
options(repr.plot.width = 20, repr.plot.height = 9)
grid.arrange(p1, p2, nrow = 1)
For choosing neighboring distance for CRAWDAD analysis, please refer to the CRAWDAD documentation.
## pkhl sample
## Define the scales to analyze the data
scales1 <- c(100, 200, 300, 400, 500, 600, 700, 800, 900, 1000)
## Shuffle cells to create null background
shuffle_list1 <- crawdad:::makeShuffledCells(cells1,
scales = scales1,
perms = 3,
ncores = 11,
seed = 1,
verbose = TRUE)
## Calculate the z-score for the cell-type pairs at different scales
results1 <- crawdad::findTrends(cells1,
neighDist = 50,
shuffleList = shuffle_list1,
ncores = 7,
verbose = TRUE,
returnMeans = FALSE)
dat_pkhl <- crawdad::meltResultsList(results1, withPerms = TRUE)
## Calculate the z-score for the multiple-test correction
zsig1 <- correctZBonferroni(dat_pkhl)
## Summary visualization
vizColocDotplot(dat_pkhl, zSigThresh = zsig1, zScoreLimit = 2*zsig1,
dotSizes = c(3,15)) +
theme(axis.text.x = element_text(angle = 35, h = 0))
## ksfb sample
## Define the scales to analyze the data
scales2 <- c(100, 200, 300, 400, 500, 600, 700, 800, 900, 1000)
## Shuffle cells to create null background
shuffle_list2 <- crawdad:::makeShuffledCells(cells2,
scales = scales2,
perms = 3,
ncores = 7,
seed = 1,
verbose = TRUE)
## Calculate the z-score for the cell-type pairs at different scales
results2 <- crawdad::findTrends(cells2,
neighDist = 50,
shuffleList = shuffle_list2,
ncores = 7,
verbose = TRUE,
returnMeans = FALSE)
dat_ksfb <- crawdad::meltResultsList(results2, withPerms = TRUE)
## Calculate the z-score for the multiple-test correction
zsig2 <- correctZBonferroni(dat_ksfb)
## Summary visualization
vizColocDotplot(dat_ksfb, zSigThresh = zsig2, zScoreLimit = 2*zsig2,
dotSizes = c(3,15)) +
theme(axis.text.x = element_text(angle = 35, h = 0))
## Add id to datasets
dat_pkhl$id <- 'pkhl'
dat_ksfb$id <- 'ksfb'
## Calculate auc
auc_samples <- calculateAUC(list(dat_pkhl, dat_ksfb))
options(repr.plot.width = 10, repr.plot.height = 9)
vizVarianceSamples(auc_samples)
reference_var <- 'Ki67 proliferating'
neighbor_var <- 'Fol B cells'
d <- dplyr::bind_rows(list(dat_ksfb, dat_pkhl))
d %>%
filter(reference == reference_var) %>%
filter(neighbor == neighbor_var) %>%
vizTrends(lines = TRUE, withPerms = TRUE, zSigThresh = zsig1)
reference_var <- 'B cells, red pulp'
neighbor_var <- 'Fol B cells'
d <- dplyr::bind_rows(list(dat_ksfb, dat_pkhl))
d %>%
filter(reference == reference_var) %>%
filter(neighbor == neighbor_var) %>%
vizTrends(lines = TRUE, withPerms = TRUE, zSigThresh = zsig1)
plot_celltypes <- function(celltypes) {
options(repr.plot.width = 18, repr.plot.height = 9)
## Combine unique cell types from both datasets
all_celltypes <- unique(c(pkhl$celltypes, ksfb$celltypes))
## Define consistent colors for all datasets
possible_colors <- c('darkblue', 'red', 'cyan', 'yellow', 'maroon')
colors <- setNames(
rep('#F0F0F0', length(all_celltypes)),
all_celltypes
)
for (i in seq_along(celltypes)) {
if (i <= length(possible_colors)) {
colors[celltypes[i]] <- possible_colors[i]
}
}
## Plot for pkhl
p1 <- pkhl %>%
mutate(celltype_highlight = celltypes %in% celltypes) %>%
arrange(celltype_highlight, celltypes) %>%
ggplot() +
geom_point(aes(x, y, color = celltypes), size = .01) +
scale_color_manual(values = colors) +
guides(colour = guide_legend(override.aes = list(size = 1))) +
labs(title = 'pkhl') +
theme_minimal()
## Plot for ksfb
p2 <- ksfb %>%
mutate(celltype_highlight = celltypes %in% celltypes) %>%
arrange(celltype_highlight, celltypes) %>%
ggplot() +
geom_point(aes(x, y, color = celltypes), size = .01) +
scale_color_manual(values = colors) +
guides(colour = guide_legend(override.aes = list(size = 1))) +
labs(title = 'ksfb') +
theme_minimal()
## Arrange plots
grid.arrange(p1, p2, nrow = 1)
}
## Example usage
reference_cell <- 'Ki67 proliferating'
neighbor_cell <- 'Fol B cells'
plot_celltypes(c(reference_cell, neighbor_cell))
reference_cell <- 'B cells, red pulp'
neighbor_cell <- 'Fol B cells'
plot_celltypes(c(reference_cell, neighbor_cell))
More details can be found in the tutorials.
CRAWDADapplied to simulated dataCRAWDADapplied to a mouse embryo seqfish dataCRAWDADapplied to a mouse cerebellum slideseq dataCRAWDADapplied to a human spleen codex data
CRAWDAD was tested on R version 4.3.0 (2023-04-21) -- "Already Tomorrow". Installation time is 1 minute in a Mac Pro Computer.