Move ggrepel / geom_text_repel's labels away from lines drawn with geom_vline() and geom_hline()

Question

ggrepel provides an excellent series of functions for annotating ggplot2 graphs and the examples page contains lots of nice hints of how to expand its functionality, including moving the labels generated away from both the axes of the plot, other labels, and so on.

However, one thing that isn't covered is moving the labels away from manually drawn lines with geom_hline() and geom_vline(), as may occur, for example, in making an annotated volcano plot.

Here's a simple MWE to highlight the problem:

library("tidyverse")
library("ggrepel")

dat <- subset(mtcars, wt > 2.75 & wt < 3.45)
dat$car <- rownames(dat)
ggplot(dat, aes(wt, mpg, label = car)) +
    geom_point(color = "red") +
    geom_text_repel(seed = 1) + #Seed for reproducibility 
    geom_vline(xintercept = 3.216) + #Deliberately chosen "bad" numbers 
    geom_hline(yintercept = 19.64) + theme_bw()

This produces the following output:

Note how the lines overlap the text of the labels and obscure it (is that "Horret 4 Drive" or "Hornet 4 Drive"?)

Jiggling the points about a bit post facto you can make a far nicer fit – I have simply shifted some of the labels a tiny bit to get them off the line.

Is it possible to get ggrepel to do this automatically? I know the example given isn't totally stable (other seeds give acceptable results) but for complex plots with a large number of points it definitely is a problem.

Edit: If you're curious, a far less "minimum" working example would be the below (taken from bioconductor):

download.file("https://raw.githubusercontent.com/biocorecrg/CRG_RIntroduction/master/de_df_for_volcano.rds", "de_df_for_volcano.rds", method="curl")
tmp <- readRDS("de_df_for_volcano.rds")
de <- tmp[complete.cases(tmp), ]
de$diffexpressed <- "NO"
# if log2Foldchange > 0.6 and pvalue < 0.05, set as "UP" 
de$diffexpressed[de$log2FoldChange > 0.6 & de$pvalue < 0.05] <- "UP"
# if log2Foldchange < -0.6 and pvalue < 0.05, set as "DOWN"
de$diffexpressed[de$log2FoldChange < -0.6 & de$pvalue < 0.05] <- "DOWN"

# Create a new column "delabel" to de, that will contain the name of genes differentially expressed (NA in case they are not)
de$delabel <- NA
de$delabel[de$diffexpressed != "NO"] <- de$gene_symbol[de$diffexpressed != "NO"]

#Actually do plot 
ggplot(data=de, aes(x=log2FoldChange, y=-log10(pvalue), col=diffexpressed, label=delabel)) +
    geom_point() + 
    theme_minimal() +
    geom_text_repel() +
    scale_color_manual(values=c("blue", "black", "red")) +
    geom_vline(xintercept=c(-0.6, 0.6), col="red") +
    geom_hline(yintercept=-log10(0.05), col="red")

This produces the below, where the text-overlapping-lines problem is quite obvious:

Answer 1

I don't think there's a built-in way to do this.

A non-elegant hack off the top of my head is to add invisible points along the intercept lines which the labels will then repel away from.

dat <- subset(mtcars, wt > 2.75 & wt < 3.45)
dat$car <- rownames(dat)

xintercept = 3.216
yintercept = 19.64

dat %>%
  mutate(alpha = 1) %>%
  bind_rows(.,
            tibble(wt = seq(from = min(.$wt), to = max(.$wt), length.out = 20), mpg = yintercept, car = '', alpha = 0),
            tibble(wt = xintercept, mpg = seq(from = min(.$mpg), to = max(.$mpg), length.out = 20), car = '', alpha = 0)
  ) %>%
  ggplot(aes(wt, mpg, label = car, alpha = alpha)) +
  geom_point(color = "red") +
  geom_text_repel(seed = 1) + #Seed for reproducibility 
  geom_vline(xintercept = xintercept) +
  geom_hline(yintercept = yintercept) + theme_bw() +
  scale_alpha_identity()