seaborn_clustermap_in_R.R

# This is a quasi replication of the correlation matrix heatmap viz from Seaborn,
# using R and ggplot
# see: http://stanford.edu/~mwaskom/software/seaborn/examples/structured_heatmap.html

library(data.table)
library(ggplot2)

# get data
brain_url <- paste0('https://raw.githubusercontent.com/',
                    'mwaskom/seaborn-data/master/brain_networks.csv')
# read data and drop 1st column, which is just row labels
brain_headers <- fread(brain_url, nrows = 3)[, -1, with=F]
brain_dat <- fread(brain_url, skip = 4)[, -1, with=F]

# collapse 3 rows of header data to single string
heads <- brain_headers[, sapply(.SD, paste, collapse='_')]
setnames(brain_dat, heads)

# Select a subset of the networks (as per example in URL)
used_networks = c(1, 5, 6, 7, 8, 11, 12, 13, 16, 17)
dat_subset <- 
  brain_dat[, as.numeric(t(brain_headers[1, ])) %in% used_networks, 
            with=F]

# corelation matrix (long format)
brain_cor <- melt(cor(dat_subset))

# reverse levels of 2nd var (to match example, though not sure why he did that;
# I think the plot is better without this)
brain_cor$Var2 <- with(brain_cor, factor(Var2, levels = rev(levels(Var2))))

ggplot(brain_cor, aes(x=Var1, y=Var2, fill=value)) +
  geom_tile(color='white', size=0.5) +
  labs(x='network - node - hemi', y='network - node - hemi') +
  scale_fill_gradient2(low='#1c545d', 
                       high = '#8a174f', 
                       mid = 'whitesmoke') +
  coord_equal() +
  theme_minimal(base_size = 10) +
  theme(axis.text.x = element_text(angle=90))
  

Obviously this plot doesn't do the hierarchical clustering + marginal dendrograms of the original example, but still a nice simple way to visualize a correlation matrix. Result: