Comparison plots for parallel coordinate vs. UMAP

Chung2017.R

library(Matrix)
library(readr)
library(tidyr)
library(dplyr)
library(ggplot2)
library(umap)
library(patchwork)
#https://www.weizmann.ac.il/sites/3CA/breast

cells <- read_csv("./data/Data_Chung2017_Breast/Cells.csv")
path <- "./data/Data_Chung2017_Breast/Exp_data_TPM.mtx"
mat <- readMM(path)
rm_wch <- which(apply(mat, MARGIN = 1, FUN = function(x)all(x==0)))

mat <- mat[-rm_wch, ]

#ca005 <- rgb(0,0,0,0.05)


system.time({
  res20_svd <- svd(log1p(as.matrix(mat)), nu = 20, nv=20)
})
# user  system elapsed 
# 15.191   0.067  15.254 

V <- res20_svd$v
colnames(V) <- paste0("PC",1:20)
df <- data.frame(V) %>% 
  mutate(cell_type=cells$cell_type, id=1:n()) %>% 
  pivot_longer(PC1:PC20) %>% 
  mutate(name = factor(name, levels=paste0("PC",1:20)))

p1 <- ggplot(df, aes(x=name,y=value,colour=cell_type, group=id))+
  geom_line(alpha=0.1, position = position_jitter(height = 0,width = 0.1, seed = 123))+
  scale_colour_brewer(palette = "Set2")+
  guides(colour=guide_legend(override.aes = list(alpha=1, linewidth=2)))+
  theme_classic()+
  ggtitle("PCA")


ures_svd <- umap(res20_svd$v)


udf_svd <- data.frame(ures_svd$layout, celltype=factor(cells$cell_type))

up2 <- ggplot(udf_svd, aes(x=X1, y=X2, colour=celltype))+
  geom_point(alpha=0.3)+
  scale_colour_brewer(palette = "Set2")+
  guides(colour=guide_legend(override.aes = list(alpha=1, linewidth=2)))+
  theme_classic()+
  ggtitle("PCA -> UMAP")+labs(x="UMAP1", y="UMAP2")

p1  / up2
ggsave("pcp_umap.png")