R ggplot heatmap with multiple rows having separate legends on the same graph

Question

I'm trying to make one heatmap using ggplot2 that contains 3 types of variables where each need their own independent legend/scale.

I am able to plot them all in one heatmap (pictured below), but I am having trouble separating them to have their own legend. My three categories are the row "Score", "samp1" and the rest of the data. I would like each of these to have their own independent legends with their respective ranges.

My only addition would be to have the row score have a green,yellow,red (low, mid,high) color scheme, if that is possible to include in this question.

This is the code I am using to create that graph

library(ggplot2)
test_data <- read.csv("test_table.csv", row.names = 1)

ggplot(test_data, aes(x=sample, y=id, fill = value)) + 
  geom_raster() + 
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1), # lables vertical
        strip.text.y = element_blank()) +  #remove facet bar on y 
  scale_fill_gradient(low = "darkblue", high = "lightblue") +
  ggtitle("test table") +
  facet_grid(rows = vars(test_data$category), 
             cols = vars(test_data$group), scales = "free", space="free_y") #facets to add gaps

I have used facets to separate the data by sample and by the 3 categories I described above. I was hoping to use this grouping to create their own legends as well, but I am not sure if this is possible.

Click here to download the data (pre-melted).

Thank you in advance.

The other approach I was considering is having them as 3 separate graphs and then trying to combine them. If that is the way to go, I could provide separated data as well. — Jakeeln, Aug 19 '20 at 19:48

score 5 · Accepted Answer · answered Aug 19 '20 at 20:03

This could be achieved via the ggnewscale package like so:

library(ggplot2)
library(dplyr)
library(ggnewscale)

ggplot() +
  geom_raster(data = filter(test_data, category == "1 score"), aes(x = sample, y = id, fill = value)) +
  scale_fill_gradient2(low = "green", mid = "yellow", high = "red", midpoint = 4, name = "Score") +
  new_scale_fill() +
  geom_raster(data = filter(test_data, category == "2 samp1"), aes(x = sample, y = id, fill = value)) +
  scale_fill_gradient(low = "darkblue", high = "lightblue", name = "Sample1") +
  new_scale_fill() +
  geom_raster(data = filter(test_data, category == "3 samp2"), aes(x = sample, y = id, fill = value)) +
  scale_fill_gradient(low = "darkblue", high = "lightblue", name = "Sample2") +
  ggtitle("test table") +
  facet_grid(
    rows = vars(category),
    cols = vars(group), scales = "free", space = "free_y"
  ) +
  theme(
    axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1),
    strip.text.y = element_blank()
  )

Thank you! This is the exact approach I was looking for. – Jakeeln Aug 19 '20 at 20:21 — Jakeeln, Aug 19 '20 at 20:21

score 3 · Answer 2 · answered Aug 19 '20 at 20:06

I would suggest next approach. Split data by groups and then build a separate plot for each group with a function. Finally use purrr and patchwork to join all plots with the diferent legends. Here the code:

library(purrr)
library(ggplot2)
library(patchwork)
#Load data
test_data <- read.csv("test_table.csv", row.names = 1)
#Split into list
List <- split(test_data,test_data$group)
#Function for plots
myfun <- function(x)
{
  G <- ggplot(x, aes(x=sample, y=id, fill = value)) + 
    geom_raster() + 
    theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1), # lables vertical
          strip.text.y = element_blank()) +  #remove facet bar on y 
    scale_fill_gradient(low = "darkblue", high = "lightblue") +
    facet_grid(rows = vars(x$category), 
               cols = vars(x$group), scales = "free", space="free_y")
  return(G)
}
#Apply
List2 <- lapply(List,myfun)
#Plot
reduce(List2, `+`)+plot_annotation(title = 'My plot')

The output:

You can explore further about patchwork and how to join multiple plots.

This is awesome, I will have to keep patchwork in mind for future projects. I really appreciate your suggestion! — Jakeeln, Aug 19 '20 at 20:20

score 1 · Answer 3 · answered Aug 19 '20 at 20:24

It's certainly possible to make 3 plots and stick them together, but given the facets are different shapes, this isn't ideal. You could use the ggnewscale package as per Stefan, which is easier now it is on CRAN, but if you want to do it in a single ggplot without add-ons it is possible. You just need to plot a geom_point that is made of large solid squares. This allows you to use a colour scale as well as a fill scale.

ggplot(test_data, aes(x=sample, y=id, fill = value)) + 
  geom_raster() + 
  geom_point(aes(alpha = id, colour = value), size = 12, shape = 15) +
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1),
        strip.text.y = element_blank()) + 
  scale_fill_gradient(low = "darkblue", high = "lightblue") +
  ggtitle("test table") +
  facet_grid(rows = vars(test_data$category), 
             cols = vars(test_data$group),  scales = "free", space = "free_y") +
  scale_alpha_manual(values = c(rep(0, 19), 1, 0, 0), guide = guide_none()) +
  scale_color_gradientn(colours = c("red", "orange", "gold", "yellow"))

R ggplot heatmap with multiple rows having separate legends on the same graph

3 Answers3

Linked