CSSN: Estimation and prediction of spatially-varying cell-specific...

In jingeyu/CSSN: Recovering spatially-varying cell-specific gene co-expression networks for single-cell spatial expression data.

Description

Functions CSSNEst and CSSNPredict in the package can provide estimates for gene co-expression networks of each cell and network predictions in a centroid location where cells are missing. CSSN can be installed in Windows, Linux, and Mac OS.

Usage

CSSNEst(X, cell.info, nu, d = 0.1, m.info = 70, is.scale = TRUE, is.all = TRUE, indx.cell = NULL, output.corr = FALSE)
CSSNPredict(GN, cell.info, miss.indx, m.info = 70)

Arguments

X	The spatial single-cell expression data matrix, where rows represent genes and columns are cells.
cell.info	The cell information matrix with dimension (n,3), where rows stand for cells, the first column represents cell types, and the second and third columns are Centroid_X and Centroid_Y coordinates, respectively.
nu	A vector with dimension n, the degrees of freedom in the Inverse-Wishart prior of Σ_i.
d	The thresholding parameter, whose range is between 0 and 1. The default value is 0.1.
m.info	The cell density parameter. The default is 70.
is.scale	a bool variable, if TRUE, the variance of gene expression data in each cell type will be scaled to 1. The default is TRUE.
is.all	bool, if TRUE, the CSSNEst returns gene co-expression networks of all cells, if FALSE, the function returns networks of assigned cells in indx.cell. The default is TRUE.
indx.cell	cell index vector, If is.all is FALSE, the function CSSNEst only calculates gene co-expression networks in indx.cell. The default is NULL.
output.corr	bool, if TRUE, the CSSNEst returns both gene co-expression networks and gene correlation matrix of prespecified cells (all cells or cells in indx.cell). if FALSE, the function only returns gene co-expression networks of prespecified cells. The default is FALSE.
GN	The estimated gene co-expression networks from the output of the function CSSNEst, a list.
miss.indx	The coordinates of the missing cells, a matrix with dimensions (miss.num, 2), where miss.num is the number of missing cells.

Value

CSSNPredict returns a list of length miss.num, each element of which is the gene co-expression network for corresponding missing cell. When parameter output.corr is FALSE, CSSNEst returns an R list with n elments, where n is the number of cells with gene co-expression networks calculated. Each element is a sparse matrix with dimension (G, G), where G is the number of genes. When output.corr is TRUE, the function returns both the gene co-expression networks and the gene correlation matrix of the cells.

References

Jinge Yu and Xiangyu Luo, "Recovering spatially-varying cell-specific gene co-expression networks for single-cell spatial expression data." (accepted by Frontiers in Genetics)

Examples

library(CSSN)
#install.packages('ggplot2')
library(ggplot2)
#install.packages('pheatmap')
library(pheatmap)

#read example data
data(example_data)

# gene number
G <- nrow(X)
# cell number
n <- ncol(X)

#---- set spatial pattern manually----
pal <- c(rgb(221, 160, 221, maxColorValue = 255),
        rgb(0, 206, 209, maxColorValue = 255))
pal <- setNames(pal, c("1", "2"))

#-----Cell's Spatial Pattern------
cell.type <- as.vector(cell.info[,1])
gg <- ggplot(cell.info, aes(x = X, y = Y, col = as.factor(cell.type), shape = as.factor(cell.type)))
pl <- gg + geom_point(size = 2.5) +
 scale_color_manual(values = c(pal[1], pal[2])) +
 theme_bw()+
 theme(legend.text=element_text(size=20),
       axis.title.x=element_text(size=16),
       axis.title.y=element_text(size=16),
       axis.text.x = element_text(size = 12,face = "bold"),
       axis.text.y = element_text(size = 12,face = "bold")
 ) + labs(x = "H", y = "L") +
 guides(color = guide_legend(title = "Cell Type",
                             title.theme = element_text(size = 25),
                             override.aes = list(size = 5)
 ),
 shape = guide_legend(title = "Cell Type",
                      title.theme = element_text(size = 25),
                      override.aes = list(size = 5)))
ggsave("cell spatial.png", pl, width = 9, height = 12)


#----run CSSNEst--------
nu <- rep(2*G, n)
Result <- CSSNEst(X, cell.info, nu = nu, d = 0.1, m.info = 70, is.scale = TRUE, is.all = TRUE)
indx.cell <- c(1,3,7,10)
result <- CSSNEst(X, cell.info, nu = nu, d = 0.1, m.info = 70, is.scale = TRUE, is.all = FALSE, indx.cell = indx.cell, output.corr = TRUE)
#-----The first five cell's estimated gene co-expression networks-----
colors_func <- colorRampPalette(c('white', "black"))
colors <- colors_func(2)
filename <- paste0("Est_", 1:5, ".png")
for(i in 1:10){
 p2 <- pheatmap(Result[[i]],
                color = colors,
                legend_breaks = c(0,1),
                cluster_cols = F, cluster_rows = F,
                show_rownames = F, show_colnames = F,
                width = 3.3, height = 2.8,
                filename = filename[i]

 )
}
# Prediction
set.seed(1)
miss.num <- 5
miss.x <- runif(miss.num, min(cell.info[,2]), max(cell.info[,2]))
miss.y <- runif(miss.num, min(cell.info[,3]), max(cell.info[,3]))
miss.indx <- cbind(miss.x, miss.y)
pre <- CSSNPredict(Result, cell.info, miss.indx)

jingeyu/CSSN documentation built on Jan. 9, 2022, 7:35 p.m.