ggplot2 multiple scales/legends per aesthetic, revisited

Question

I have an example where I'd like to highlight several properties of sequence alignments using ggplot. I'm using geom_tile and want to have two sets of differently coloured tiles for two score properties. I am only able to visualize one.

I am aware of the limitation of one scale per aesthetic (and the logic behind it), but maybe someone has an idea how to hack it for cases like this where it would make sense to have different colour scales in one 'plot'.

Perhaps with adding manually the Grobs, but I wouldn't know where to start...

an additional question: for some reason the override.aes=list(shape = "A") does not work, any ideas why?

one more: any method to scale text proportionally to the size of the tile (or the other way around)?

library(ggplot2)
library(grid)

pd = data.frame(
  letters = strsplit("AGTGACCGACTATCATAGTGACCCAGAATCATAGTGACCGAGTATGAT", "")[[1]],
  species = rep(c("Human", "Armadillo", "Porcupine"), each=16),
  x       = rep(1:16, 3),
  change  = c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
              0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,
              0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0),
  score1  = c(0,0,0,0,0,0,1,1,2,2,2,3,3,3,4,3,
              0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,
              0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
  score2  = c(0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,
              0,0,0,0,2,2,2,2,0,0,0,0,0,0,0,0,
              0,0,0,0,3,3,3,3,0,0,0,0,0,0,0,0)
)


ggplot(pd[pd$score1 != 0,], aes(x=x, y=species)) +
  coord_fixed(ratio = 1.5, xlim=c(0.5,16.5), ylim=c(0.5, 3.5)) +
  geom_tile(aes(fill=score1)) +
  scale_fill_gradient2("Score 1", limits=c(0,4),low="#762A83", mid="white", high="#1B7837", guide=guide_colorbar(title.position="top")) +
  geom_text(data=pd, aes(label=letters, color=factor(change)), size=rel(5), family="mono") +
  scale_color_manual("Change", values=c("black", "#F2A11F"), labels=c("None", "Some"), guide=guide_legend(direction="vertical", title.position="top", override.aes=list(shape = "A"))) +
  theme(panel.background=element_rect(fill="white", colour="white"),
        axis.title = element_blank(),
        axis.ticks.y = element_blank(),
        axis.text.y = element_text(family="mono", size=rel(2)),
        axis.text.x = element_text(size=rel(0.7)),
        legend.text = element_text(size=rel(0.7)),
        legend.key.size = unit(0.7, "lines"),
        legend.position = "bottom", legend.box = "horizontal") +
  ggtitle("What about Score2?")

How to add another layer of tiles with different color scale?

score 9 · Accepted Answer · answered Jun 08 '13 at 18:35

I managed to get a satisfactory result by combining grobs from two separately generated plots. I'm sure the solution can be generalized better to accommodate different grob indices ...

library(ggplot2)
library(grid)

pd = data.frame(
  letters = strsplit("AGTGACCGACTATCATAGTGACCCAGAATCATAGTGACCGAGTATGAT", "")[[1]],
  species = rep(c("Human", "Armadillo", "Porcupine"), each=16),
  x       = rep(1:16, 3),
  change  = c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
              0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,
              0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0),
  score1  = c(0,0,0,0,0,0,1,1,2,2,2,3,3,3,4,3,
              0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,
              0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
  score2  = c(0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,
              0,0,0,0,2,2,2,2,0,0,0,0,0,0,0,0,
              0,0,0,0,3,3,3,3,0,0,0,0,0,0,0,0)
)


p1=ggplot(pd[pd$score1 != 0,], aes(x=x, y=species)) +
  coord_fixed(ratio = 1.5, xlim=c(0.5,16.5), ylim=c(0.5, 3.5)) +
  geom_tile(aes(fill=score1)) +
  scale_fill_gradient2("Score 1", limits=c(0,4),low="#762A83", mid="white", high="#1B7837", guide=guide_colorbar(title.position="top")) +
  geom_text(data=pd, aes(label=letters, color=factor(change)), size=rel(5), family="mono") +
  scale_color_manual("Change", values=c("black", "#F2A11F"), labels=c("None", "Some"), guide=guide_legend(direction="vertical", title.position="top", override.aes=list(shape = "A"))) +
  theme(panel.background=element_rect(fill="white", colour="white"),
        axis.title = element_blank(),
        axis.ticks.y = element_blank(),
        axis.text.y = element_text(family="mono", size=rel(2)),
        axis.text.x = element_text(size=rel(0.7)),
        legend.text = element_text(size=rel(0.7)),
        legend.key.size = unit(0.7, "lines"),
        legend.position = "bottom", legend.box = "horizontal") +
  ggtitle("Voila, the Score2!")

p2=ggplot(pd[pd$score2 != 0,], aes(x=x, y=species)) +
  coord_fixed(ratio = 1.5, xlim=c(0.5,16.5), ylim=c(0.5, 3.5)) +
  geom_tile(aes(fill=score2)) +
  scale_fill_gradient2("Score 2", limits=c(0,3),low="#1B7837", mid="white", high="#762A83", guide=guide_colorbar(title.position="top")) +
  geom_text(data=pd, aes(label=letters, color=factor(change)), size=rel(5), family="mono") +
  scale_color_manual("Change", values=c("black", "#F2A11F"), labels=c("None", "Some"), guide=guide_legend(direction="vertical", title.position="top", override.aes=list(shape = "A"))) +
  theme(panel.background=element_rect(fill="white", colour="white"),
        axis.title = element_blank(),
        axis.ticks.y = element_blank(),
        axis.text.y = element_text(family="mono", size=rel(2)),
        axis.text.x = element_text(size=rel(0.7)),
        legend.text = element_text(size=rel(0.7)),
        legend.key.size = unit(0.7, "lines"),
        legend.position = "bottom", legend.box = "horizontal") +
  ggtitle("What about Score2?")


p1g=ggplotGrob(p1)
p2g=ggplotGrob(p2)

combo.grob = p1g

combo.grob$grobs[[8]] = cbind(p1g$grobs[[8]][,1:4], 
                              p2g$grobs[[8]][,3:5], 
                              size="first")

combo.grob$grobs[[4]] = reorderGrob(
                          addGrob(p1g$grobs[[4]], 
                                  getGrob(p2g$grobs[[4]], 
                                          "geom_rect.rect", 
                                          grep=TRUE)), 
                          c(1,2,5,3,4))
grid.newpage()
grid.draw(combo.grob)

Two scales in one plot

kohske · Answer 2 · 2013-04-23T00:54:33.190

I would use size of text for indicating the score2:

ggplot(pd[pd$score1 != 0,], aes(x=x, y=species)) +
  coord_fixed(ratio = 1.5, xlim=c(0.5,16.5), ylim=c(0.5, 3.5)) +
  geom_tile(aes(fill=score1)) +
  scale_fill_gradient2("Score 1", limits=c(0,4),low="#762A83", mid="white", high="#1B7837", guide=guide_colorbar(title.position="top")) +
  geom_text(data=pd, aes(label=letters, size = score2, color=factor(change)), family="mono") +
  scale_size_continuous(range = c(4, 8)) +
  scale_color_manual("Change", values=c("black", "#F2A11F"), labels=c("None", "Some"), guide=guide_legend(direction="vertical", title.position="top", override.aes=list(shape = "A"))) +
  theme(panel.background=element_rect(fill="white", colour="white"),
        axis.title = element_blank(),
        axis.ticks.y = element_blank(),
        axis.text.y = element_text(family="mono", size=rel(2)),
        axis.text.x = element_text(size=rel(0.7)),
        legend.text = element_text(size=rel(0.7)),
        legend.key.size = unit(0.7, "lines"),
        legend.position = "bottom", legend.box = "horizontal") +
  ggtitle("What about Score2?")

enter image description here

UPDATED:

Here is a quick hack, I'm not sure if this is easy to inspect visually though...

library(ggplot2)
library(grid)
library(proto)

GeomTile2 <- proto(ggplot2:::GeomTile, {
  reparameterise <- function(., df, params) {
    df <- .$.super$reparameterise(df, params)
    if (params$ud == "u") 
      transform(df, ymin = y) 
    else 
        transform(df, ymax = (y-ymin)*0.8 + ymin, ymin = (y-ymin)*0.2 + ymin)
  }
  draw <- function(..., ud) {.$.super$draw(..., ud)}
})
geom_tile2 <- function (mapping = NULL, data = NULL, stat = "identity", position = "identity", ..., ud = "u") { 
  GeomTile2$new(mapping = mapping, data = data, stat = stat, position = position, ..., ud = ud)
}

ggplot(pd, aes(x=x, y=species)) +
  coord_fixed(ratio = 1.5, xlim=c(0.5,16.5), ylim=c(0.5, 3.5)) +
  geom_tile2(aes(fill=score1), ud = "u") +
  geom_tile2(aes(fill = score2), ud = "d") +
  scale_fill_gradient2("Score 1", limits=c(0,4),low="#762A83", mid="white", high="#1B7837", guide=guide_colorbar(title.position="top")) +
  geom_text(data=pd, aes(label=letters, color=factor(change)), size=rel(5), family="mono") +
  scale_color_manual("Change", values=c("black", "#F2A11F"), labels=c("None", "Some"), guide=guide_legend(direction="vertical", title.position="top", override.aes=list(shape = "A"))) +
  theme(panel.background=element_rect(fill="white", colour="white"),
        axis.title = element_blank(),
        axis.ticks.y = element_blank(),
        axis.text.y = element_text(family="mono", size=rel(2)),
        axis.text.x = element_text(size=rel(0.7)),
        legend.text = element_text(size=rel(0.7)),
        legend.key.size = unit(0.7, "lines"),
        legend.position = "bottom", legend.box = "horizontal") +
  ggtitle("What about Score2?")

enter image description here

Upper half indicates score1 while the lower for score2.

Thanks koshke, I have considered that, but it is not a sufficient visual cue to represent this property. These alignments can be a hundred or more characters in length and and can contain several dozens of species, so the letters are fairly small as it is and there is often no space to manipulate their size. Moreover, sometimes the score is a continuous variable and it will be really hard to discern the relative quantities from one sequence to the other. I'm afraid I'd need something more 'visual' — Krizbi, Apr 22 '13 at 06:03
Just to give a more real example: [here is a complete alignment](http://i.imgur.com/wB4prkv.png) — Krizbi, Apr 22 '13 at 06:11
Hi @kohske thanks for the effort, but I'm afraid it does not represent visually enough distinct information. I'd need to have something complementary in color because it is a whole different property. Plus, the legend is also something that very quickly tells the observer how many different properties are there in one figure. So, scale is the key, not the geom. The best I could come up is involving alpha channel and mapping it to a line type, but this has a downside of having to deal with a single continuous color ramp. [An example is here](http://imgur.com/EOLXg3J) — Krizbi, Apr 23 '13 at 06:18
Perhaps a suggestion to @hadley to consider: a compromise from "one aes - one scale" rule would be to allow only annotation geoms to have unlimited combinations of scales and aesthetics. An annotation would then be as "colourful" as possible but without interfering with the actual (and very sensible) rule and have their own scale for as many annotation layers there are. This way, plots could be annotated with more creative freedom by the user and they'd keep all the nice features and visual consistency of ggplot2. Just a thought... — Krizbi, Apr 23 '13 at 06:26

ggplot2 multiple scales/legends per aesthetic, revisited

2 Answers2

Linked

Related