R ggplot2: boxplots with wilcoxon significance levels, and facets. Show only significant comparisons with asterisks

Question

Following up on this question and for the sake of completeness, I modified the accepted answer and customized the resulting plot, but I am still facing some important proble

Answer 1

You can try following. As your code is really busy and for me too complicated to understand, I suggest a different approach. I tried to avoid loops and to use the tidyverse as much as possible. Thus, first I created your data. Then calculated kruskal wallis tests as this was not possible within ggsignif. Afterwards I will plot all p.values using geom_signif. Finally, insignificant ones will be removed and a step increase is added.

1- Make coloring work done

2- Show asterisks instead of numbers done

...and for the win:

3- Make a common legend done

4- Place Kruskal-Wallis line on top done, I placed the values at the bottom

5- Change the size (and alignment) of the title and y axis text done

library(tidyverse)
library(ggsignif)

# 1. your data
set.seed(2)
df <- as.tbl(iris) %>% 
  mutate(treatment=rep(c("A","B"), length(iris$Species)/2)) %>% 
  gather(key, value, -Species, -treatment) %>% 
  mutate(value=rnorm(n())) %>% 
  mutate(key=factor(key, levels=unique(key))) %>% 
  mutate(both=interaction(treatment, key, sep = " "))

# 2. Kruskal test
KW <- df %>% 
  group_by(Species) %>%
  summarise(p=round(kruskal.test(value ~ both)$p.value,2),
            y=min(value),
            x=1) %>% 
  mutate(y=min(y))

# 3. Plot  
P <- df %>% 
ggplot(aes(x=both, y=value)) + 
  geom_boxplot(aes(fill=Species)) + 
  facet_grid(~Species) +
  ylim(-3,7)+
  theme(axis.text.x = element_text(angle=45, hjust=1)) +
  geom_signif(comparisons = combn(levels(df$both),2,simplify = F),
              map_signif_level = T) +
  stat_summary(fun.y=mean, geom="point", shape=5, size=4) +
  xlab("") +
  geom_text(data=KW,aes(x, y=y, label=paste0("KW p=",p)),hjust=0) +
  ggtitle("Plot") + ylab("This is my own y-lab")

# 4. remove not significant values and add step increase
P_new <- ggplot_build(P)
P_new$data[[2]] <- P_new$data[[2]] %>% 
  filter(annotation != "NS.") %>% 
  group_by(PANEL) %>%
  mutate(index=(as.numeric(group[drop=T])-1)*0.5) %>% 
  mutate(y=y+index,
         yend=yend+index) %>% 
  select(-index) %>% 
  as.data.frame()
# the final plot  
plot(ggplot_gtable(P_new))

and similar approach using two facets

# --------------------
# 5. Kruskal
KW <- df %>% 
  group_by(Species, treatment) %>%
  summarise(p=round(kruskal.test(value ~ both)$p.value,2),
            y=min(value),
            x=1) %>% 
  ungroup() %>% 
  mutate(y=min(y))


# 6. Plot with two facets  
P <- df %>% 
  ggplot(aes(x=key, y=value)) + 
  geom_boxplot(aes(fill=Species)) + 
  facet_grid(treatment~Species) +
  ylim(-5,7)+
  theme(axis.text.x = element_text(angle=45, hjust=1)) +
  geom_signif(comparisons = combn(levels(df$key),2,simplify = F),
              map_signif_level = T) +
  stat_summary(fun.y=mean, geom="point", shape=5, size=4) +
  xlab("") +
  geom_text(data=KW,aes(x, y=y, label=paste0("KW p=",p)),hjust=0) +
  ggtitle("Plot") + ylab("This is my own y-lab")

# 7. remove not significant values and add step increase
P_new <- ggplot_build(P)
P_new$data[[2]] <- P_new$data[[2]] %>% 
  filter(annotation != "NS.") %>% 
  group_by(PANEL) %>%
  mutate(index=(as.numeric(group[drop=T])-1)*0.5) %>% 
  mutate(y=y+index,
         yend=yend+index) %>% 
  select(-index) %>% 
  as.data.frame()
# the final plot  
plot(ggplot_gtable(P_new))

Edit.

Regarding to your p.adjust needs, you can set up a function on your own and calling it directly within geom_signif().

wilcox.test.BH.adjusted <- function(x,y,n){
  tmp <- wilcox.test(x,y)
  tmp$p.value <- p.adjust(tmp$p.value, n = n,method = "BH")
  tmp
}  

geom_signif(comparisons = combn(levels(df$both),2,simplify = F),
          map_signif_level = T, test = "wilcox.test.BH.adjusted", 
          test.args = list(n=8))

The challenge is to know how many independet tests you will have in the end. Then you can set the n by your own. Here I used 8. But this is maybe wrong.

Answer 2

Constructing ggplots in a loop has always been known to produce confusing results, and for the explanation of point 1 I'll refer to this question and many others. There's also a hint there about evaluating the ggplot object on the spot, e.g. via print. Re point 2, you were close, a bit of debugging with trial and error helped. Here's the complete code for plot.list:

plot.list=function(mydf, pv.final, addkw, a, myPal){
    mylist <- list()
    i <- 0
    for (sp in unique(mydf$Species)){
        i <- i+1
        mydf0 <- subset(mydf, Species==sp)
        addkw0 <- subset(addkw, Species==sp)
        pv.final0 <- pv.final[grep(sp, pv.final$group1), ]
        num.signif <- sum(pv.final0$p.value <= 0.05)
        P <- ggplot(mydf0,aes(x=both, y=value)) +
            geom_boxplot(aes(fill=Species)) +
            stat_summary(fun.y=mean, geom="point", shape=5, size=4) +
            facet_grid(~Species, scales="free", space="free_x") +
            scale_fill_manual(values=myPal[i]) +
            geom_text(data=addkw0, hjust=0, size=4.5, aes(x=0, y=round(max(mydf0$value, na.rm=TRUE)+0.5), label=paste0("KW p=",p.value))) +
            geom_signif(test="wilcox.test", comparisons = a[which(pv.final0$p.value<=0.05)],#I can use "a"here
                        map_signif_level = F,            
                        vjust=0,
                        textsize=4,
                        size=0.5,
                        step_increase = 0.05)
        if (i==1){
            P <- P + theme(legend.position="none",
                           axis.text.x=element_text(size=20, angle=90, hjust=1),
                           axis.text.y=element_text(size=20),
                           axis.title=element_blank(),
                           strip.text.x=element_text(size=20,face="bold"),
                           strip.text.y=element_text(size=20,face="bold"))
        } else{
            P <- P + theme(legend.position="none",
                           axis.text.x=element_text(size=20, angle=90, hjust=1),
                           axis.text.y=element_blank(),
                           axis.ticks.y=element_blank(),
                           axis.title=element_blank(),
                           strip.text.x=element_text(size=20,face="bold"),
                           strip.text.y=element_text(size=20,face="bold"))
        }
        P2 <- ggplot_build(P)
        P2$data[[4]]$annotation <- rep(subset(pv.final0, p.value<=0.05)$map.signif, each=3)
        P <- ggplot_gtable(P2)
        mylist[[sp]] <- list(num.signif, P)
    }
    return(mylist)
}

Note that we can no longer modify the plot via ggplot semantics, since we already applied ggplot_build/ggplot_gtable, so scale modification is no longer possible. If you want to preserve it, move it inside the plot.list function. So, changing to

grid.arrange(grobs=lapply(p.list, function(x) x[[2]]), 
             ncol=length(unique(mydf$Species)), top="Random title", left="Value")

yields

That's not a complete solution, of course, but I hope that helps.