Dynamic midpoint in ggplot2's scale_fill_gradient2 - r

Question

I'm making a heatmap in R using ggplot2 and I want to dynamically change the value of midpoint for scale_fill_gradient2. I want the midpoint for every row to be the maximum of v1 and v2.
Here's the original plot and data:
library(ggplot2)
set.seed(1L)
s = sprintf("d%s", 1:9)
vars = sprintf("v%s", 1:6)
data = data.frame(s = rep(s, 6), stringsAsFactors = FALSE)
data$variable = rep(vars, rep.int(9, 6))
data$variable = as.factor(data$variable)
data$value = round(runif(54, min=-100, max=100), 1)
pdf(save)
heatmap = ggplot(data = data, aes(x = variable, y = s, fill = value)) +
geom_tile(color = "black", aes(width = 1)) +
scale_fill_gradient2(low = cbbPalette$pink, high = cbbPalette$green, mid = cbbPalette$grey,
midpoint = 0, space = "Lab",
name = title) +
scale_color_discrete("exps", data$variable) +
theme_minimal() +
theme(axis.text.x = element_text(vjust = 1,
size = title.size), legend.title = element_blank(),
axis.text.y = element_text(size = title.size),
strip.text.x = element_text(size = title.size)) +
coord_fixed()
#add numbers to cells
heatmap = heatmap + geom_text(aes(x = variable, y = s, label = value), color = cbbPalette$black, size = 3) +
theme(
axis.title.x = element_blank(),
axis.title.y = element_blank(),
panel.grid.major = element_blank(),
panel.border = element_blank(),
panel.background = element_blank(),
axis.ticks = element_blank(),
legend.justification = c(0.5, 0),
legend.direction = "horizontal",
legend.position = "top") +
guides(fill = guide_colorbar(barwidth = 7, barheight = 1,
title.position = "top", title.hjust = 0.5))
# Print the heatmap
print(heatmap)
dev.off()
I tried to change midpoint by taking max of v1 and v2 but that affects all rows instead each row separately.
scale_fill_gradient2(low = cbbPalette$pink, high = cbbPalette$green, mid = cbbPalette$grey,
midpoint = data[data$variable == "v1", "value"], space = "Lab",
name = title)

Answer 1

Scales don't really work that way, as they map a range of values to a set of colours. Consequentially, a particular colour means a particular value for the whole plot. My best advice would be to pre-normalise the data by subtracting the max of v1/v2. See example in code below (there were a few variables in your example but not in the shared code which I've subsituted).
library(ggplot2)
library(tidyverse)
set.seed(1L)
s = sprintf("d%s", 1:9)
vars = sprintf("v%s", 1:6)
data = data.frame(s = rep(s, 6), stringsAsFactors = FALSE)
data$variable = rep(vars, rep.int(9, 6))
data$variable = as.factor(data$variable)
data$value = round(runif(54, min=-100, max=100), 1)
new_data <- data %>% group_by(s) %>%
mutate(value = value - max(value[variable %in% c("v1", "v2")]))
ggplot(data = new_data, aes(x = variable, y = s, fill = value)) +
geom_tile(color = "black", aes(width = 1)) +
scale_fill_gradient2(low = "pink", high = "green", mid = "grey",
midpoint = 0, space = "Lab",
name = "title") +
scale_color_discrete("exps", data$variable) +
theme_minimal() +
coord_fixed()