Constellation plot

Creating a constellation plot

This notebook can be used to create a constellation plot.

For the purpose of this notebook, we will use the data from GABAergic CGE-derived neurons in cortex. This data is a subset of the data from the Allen Brain Cell Atlas.

This notebook will first construct the knn matrix by running the get_knn_graph() function which serves as part of the input for the plot_constellation() function. More information about the functions and their input requirements can be found here.

library(ggplot2)
library(purrr)
library(scrattch.vis)
library(scrattch.hicat)
library(Matrix)
library(dplyr)
library(tidyr)
library(parallel)
library(bigstatsr)
library(scrattch.bigcat)
library(data.table)
library(arrow)
library(BiocNeighbors)

URL = 'https://allen-brain-cell-atlas.s3-us-west-2.amazonaws.com/metadata/WMB-taxonomy/20231215/cl.df_CCN202307220.xlsx'
data = rio::import_list(URL)

colors <- rio::import("https://allen-brain-cell-atlas.s3-us-west-2.amazonaws.com/metadata/WMB-taxonomy/20231215/views/cluster_to_cluster_annotation_membership_color.csv")

URL = "https://allen-brain-cell-atlas.s3-us-west-2.amazonaws.com/metadata/WMB-10X/20231215/cell_metadata.csv"
anno.df = rio::import(URL)

cl.df <- data$cluster_annotation
cl.df <- cl.df[cl.df$class_label != "LQ",]

# add colors to cluster data frame
colors$cluster_alias <- as.character(as.integer(colors$cluster_alias))
cl.df <- cl.df %>% left_join(colors, by=c("cl"="cluster_alias"))

select.columns <- colnames(cl.df)[grep("^supertype", colnames(cl.df))]
st.df <- cl.df %>% group_by_at(select.columns) %>% summarise(n=n())

select.columns <- colnames(cl.df)[grep("^subclass", colnames(cl.df))]
sc.df <- cl.df %>% group_by_at(select.columns) %>% summarise(n=n())

select.columns <- colnames(cl.df)[grep("^class", colnames(cl.df))]
c.df <- cl.df %>% group_by_at(select.columns) %>% summarise(n=n())

cl.df <- cl.df %>% filter(class_id == 6)
cl.df$cl <- as.integer(as.character(cl.df$cl))

anno.df <- anno.df %>% 
  filter(cluster_alias %in% cl.df$cl) %>% 
  # filter(library_method == "10Xv3") %>%
  left_join(cl.df[,c(1:12)], by=c("cluster_alias"="cl")) 
rownames(anno.df) <- anno.df$cell_label

One of the inputs required to generate the KNN graph is a reduced dimension matrix. For the purpose of this notebook a pre-made matrix is downloaded

URL <- "https://allen-share.s3.us-east-2.amazonaws.com/scrattch.example/CTX-CGE_GABA.rd.dat.csv"

rd.dat.df <- rio::import(URL)
rd.dat = as.matrix(rd.dat.df[,-1])
row.names(rd.dat) = rd.dat.df[[1]]

URL <- "https://allen-share.s3.us-east-2.amazonaws.com/scrattch.example/CTX-CGE_GABA.umap.2d.csv"

umap.df <- rio::import(URL, header = T)
row.names(umap.df) = umap.df[[1]] 
umap.df = umap.df[,-1]

Cluster constellation

cl.cl = with(anno.df %>% filter(library_method=="10Xv3" & 
  !is.na(subclass_id_label) & cluster_alias %in% cl.df$cl), setNames(cluster_id, cell_label))

rd.dat = rd.dat[intersect(rownames(rd.dat),names(cl.cl)),]

result = get_knn_graph(rd.dat, cl=cl.cl, k =50) 

Select the most robust edges for plotting

knn.cl.df = result$knn.cl.df 
knn.cl.df = knn.cl.df %>% group_by(cl.from) %>% mutate(cl.from.rank = rank(-Freq))
knn.cl.df = knn.cl.df %>% group_by(cl.to) %>% mutate(cl.to.rank = rank(-Freq))
select.knn.cl.df = with(knn.cl.df, knn.cl.df[odds > 1 & pval.log < log(1/100000) & (frac > 0.1 | frac > 0.03 & Freq > 100) & (cl.from.rank < 4| cl.to.rank < 4),])

Create the nodes to be plotted on the constellation. The nodes are located at the centroid location of cells within a grouping of cells. The grouping of cells in this case is cluster. For plotting a few columns are expected by the function, these include;

• cluster_id

• cluster_label

• cluster_color

• cluster_size

prefix = "CTX-CGE"

cl.center.df = as.data.frame(get_RD_cl_center(umap.df,cl.cl)) 

cl.df$cluster_size <- rowSums(cl.df[,c("v3.size", "v2.size", "multiome.size"   )])

cl.center.df$cluster_id = as.integer(as.character(row.names(cl.center.df) ))
cl.center.df <- cl.center.df %>% left_join(cl.df[,c(1:12,48:53)])
cl.center.df$cluster_label <- cl.center.df$cluster_id_label
rownames(cl.center.df) <- cl.center.df$cluster_id

# set cl as cluster_id since that was used to summarise the edges
cl.center.df$cl = as.integer(as.character(row.names(cl.center.df) ))

tmp.cl = cl.center.df$cluster_id
tmp.knn.cl.df = select.knn.cl.df %>% filter(cl.from %in% tmp.cl & cl.to %in% tmp.cl)

  c.plot=plot_constellation(tmp.knn.cl.df, 
                            cl.center.df=cl.center.df, 
                            out.dir="./",
                            prefix=paste0(prefix,".cluster.stcol"),
                            node.label="cluster_id",
                            exxageration=2,
                            plot.parts=FALSE,
                            return.list = T,
                            node.dodge = F,
                            label_repel = TRUE,
                            label.size = 4,
                            plot.height = 15,
                            plot.width = 15)
c.plot$constellation

Adjust the color of the nodes to reflect subclasses

cl.center.df$cluster_color <- cl.center.df$subclass_color
  c.plot=plot_constellation(tmp.knn.cl.df, 
                            cl.center.df=cl.center.df, 
                            out.dir="./", # directory for saving files
                            prefix=paste0(prefix,".cluster.sccol"), #filename for saving
                            node.label="cluster_id",
                            exxageration=2,
                            plot.parts=FALSE,
                            return.list = T,
                            node.dodge = F,
                            label_repel = TRUE,
                            label.size = 4,
                            plot.height = 15,
                            plot.width = 15)
c.plot$constellation