ggplot heatmap gridline formatting geom_tile and geom_rect

Question

I have been working on creating a heatmap for a few days and I cannot get the final formating of gridlines to work. See the codes and attached plots below. What I am trying to do is to align the gridline along the tiles of the heatmap using geom_tile() so each tile fills the inside of the grid in a box way. I was able to align the gridlines using geom_raster() but the y-axis label ticks at either the top or the bottom of the tile but I need it to tick at the center (See red highlight), also I cannot get geom_raster to wrap a white line border around the tiles so the color blocks looks a bit disorganized in my original dataset. Would be grateful for any help with the formatting codes. Thanks very much!

#The data set in long format 


y<- c("A","A","A","A","B","B","B","B","B","C","C","C","D","D","D")
    x<- c("2020-03-01","2020-03-15","2020-03-18","2020-03-18","2020-03-01","2020-03-01","2020-03-01","2020-03-01","2020-03-05","2020-03-06","2020-03-05","2020-03-05","2020-03-20","2020-03-20","2020-03-21")
    v<-data.frame(y,x)

#approach 1 using geom_tile but gridline does not align with borders of the tiles 
    v%>%
      count(y,x,drop=FALSE)%>%
      arrange(n)%>%
      ggplot(aes(x=x,y=fct_reorder(y,n,sum)))+
      geom_tile(aes(fill=n),color="white", size=0.25)

I have tried running similar codes from another post but I wasn't able to get it to run properly. I think because my x variable is a count variable of y variable so cannot be formatted into a factor variable to specify xmin and xmax in geom_rect()

#approach 2 using geom_raster but y-axis label can't tick at the center of tiles and there's no border around the tile to differentiate between tiles. 

v%>%
  count(y,x,drop=FALSE)%>%
  arrange(n)%>%
  ggplot()+
  geom_raster(aes(x=x,y=fct_reorder(y,n,sum),fill=n),hjust=0,vjust=0)

score 3 · Answer 1 · answered May 23 '20 at 08:28

I think it makes sense to keep the ticks and in turn the grid lines where they are. To still achieve what you're looking for, I would suggest you expand your data to include all possible combinations and simply set the na.value to a neutral fill color:

# all possible combinations
all <- v %>% expand(y, x)

# join with all, n will be NA for obs. in all that are not present in v
v = v %>% group_by_at(vars(y, x)) %>% 
    summarize(n = n()) %>% right_join(all)

ggplot(data = v, 
       aes(x=x, y=fct_reorder(y,n, function(x) sum(x, na.rm = T))))+ # note that you must account for the NA values now 
geom_tile(aes(fill=n), color="white",
        size=0.25) +
scale_fill_continuous(na.value = 'grey90') +
scale_x_discrete(expand = c(0,0)) +
scale_y_discrete(expand = c(0,0))

Thanks PinotTiger. The codes ran well in the sample dataset (v) However, when the codes ran in my actual dataset, the y gridline is still passing through the middle of the tiles from the ticks line of y labels which is not what i want in the plot. — DHR, May 23 '20 at 10:20
Mhh, difficult to tell what the issue is without the actual dataset... — PRZ, May 23 '20 at 14:08

stefan · Accepted Answer · 2020-05-23T10:22:18.627

This is a bit of a hack. My approach converts the categorical variables to numerics which adds minor grid lines to the plot which align with the tiles. To get rid of the major grid lines I simply use theme(). Drawback: Breaks and labels have to be set manually.

library(ggplot2)
library(dplyr)
library(forcats)

v1 <- v %>%
  count(y,x,drop=FALSE)%>%
  arrange(n) %>%
  mutate(y = fct_reorder(y, n, sum),
         y1 = as.integer(y),
         x = factor(x),
         x1 = as.integer(x))

labels_y <- levels(v1$y)
breaks_y <- seq_along(labels_y)

labels_x <- levels(v1$x)
breaks_x <- seq_along(labels_x)

ggplot(v1, aes(x=x1, y=y1))+
  geom_tile(aes(fill=n), color="white", size=0.25) + 
  scale_y_continuous(breaks = breaks_y, labels = labels_y) +
  scale_x_continuous(breaks = breaks_x, labels = labels_x) +
  theme(panel.grid.major = element_blank())

^{Created on 2020-05-23 by the reprex package (v0.3.0)}

Edit: Checked for long var names

y<- c("John Doe","John Doe","John Doe","John Doe","Mary Jane","Mary Jane","Mary Jane","Mary Jane","Mary Jane","C","C","C","D","D","D")
x<- c("2020-03-01","2020-03-15","2020-03-18","2020-03-18","2020-03-01","2020-03-01","2020-03-01","2020-03-01","2020-03-05","2020-03-06","2020-03-05","2020-03-05","2020-03-20","2020-03-20","2020-03-21")
v<-data.frame(y,x)

^{Created on 2020-05-23 by the reprex package (v0.3.0)}

Thanks, Stefan. However, my y variable in the actual data set had longer character labels (e.g. "Mary Jane", "John Doe"...etc) and therefore cannot convert to integers and appeared as NA as y1. What would you advise in this case. My apologies I should have created the sample dataset with that in mind. — DHR, May 23 '20 at 10:12
Hi @DHR. Have you converted to factors before converting to numerics? Have a look at my edit. Changed A to John Doe and B to Mary Jane. My code still works for that case. — stefan, May 23 '20 at 10:24

ggplot heatmap gridline formatting geom_tile and geom_rect

2 Answers2