如何在R分组柱状图中调整变量标签格式并按链接方法分组？

现有如下data.table格式的数据：

dataset            variable       ARI
 1:    pcaZ0     pearson, single 0.6984690
 2:    pcaZ0   pearson, complete 0.6984690
 3:    pcaZ0    pearson, average 0.6984690
 4:    pcaZ0    spearman, single 0.6984690
 5:    pcaZ0  spearman, complete 0.6984690
 6:    pcaZ0   spearman, average 0.6984690
 7:    pcaZ0      cosine, single 0.7238611
 8:    pcaZ0    cosine, complete 0.7238611
 9:    pcaZ0     cosine, average 0.7109783
10:    pcaZ0   euclidean, single 0.5177371
11:    pcaZ0 euclidean, complete 0.5177371
12:    pcaZ0  euclidean, average 0.5177371
13:    pcaZ1     pearson, single 0.5429425
14:    pcaZ1   pearson, complete 0.9619119
15:    pcaZ1    pearson, average 0.5429425
16:    pcaZ1    spearman, single 0.5317401
17:    pcaZ1  spearman, complete 0.5317401
18:    pcaZ1   spearman, average 0.7371173
19:    pcaZ1      cosine, single 0.5220314
20:    pcaZ1    cosine, complete 0.9434279
21:    pcaZ1     cosine, average 0.8089993
22:    pcaZ1   euclidean, single 0.5177371
23:    pcaZ1 euclidean, complete 0.5177371
24:    pcaZ1  euclidean, average 0.5177371
25: modpcaZ0     pearson, single 0.5251420
26: modpcaZ0   pearson, complete 0.8485167
27: modpcaZ0    pearson, average 0.9596045
28: modpcaZ0    spearman, single 0.5251420
29: modpcaZ0  spearman, complete 0.8485167
30: modpcaZ0   spearman, average 0.8838628
31: modpcaZ0      cosine, single 0.5083105
32: modpcaZ0    cosine, complete 0.9596045
33: modpcaZ0     cosine, average 0.9596045
34: modpcaZ0   euclidean, single 0.5203030
35: modpcaZ0 euclidean, complete 0.6717862
36: modpcaZ0  euclidean, average 0.5360825
37: modpcaZ1     pearson, single 0.5360825
38: modpcaZ1   pearson, complete 0.8485167
39: modpcaZ1    pearson, average 0.8838628
40: modpcaZ1    spearman, single 0.5630128
41: modpcaZ1  spearman, complete 0.8485167
42: modpcaZ1   spearman, average 0.8485167
43: modpcaZ1      cosine, single 0.5360825
44: modpcaZ1    cosine, complete 0.8314749
45: modpcaZ1     cosine, average 0.9400379
46: modpcaZ1   euclidean, single 0.5360825
47: modpcaZ1 euclidean, complete 0.7239638
48: modpcaZ1  euclidean, average 0.5487061

使用以下代码生成分组柱状图：

library(tidyverse)
library(ggplot2)
library(gridExtra)
library(ggtext)
library(khroma)

# grouped bar charts
ggplot(b2, aes(y=variable, x=ARI, fill=dataset)) + 
geom_col(position=position_dodge(), width=0.6) +
scale_fill_manual(name=NULL,
                  breaks=c("pcaZ0", "pcaZ1", "modpcaZ0", "modpcaZ1"),
                  labels=c("non-standard.", 
                           "standard.", 
                           "non-standard.<br>*without 71-76*",
                           "standard.<br>*without 71-76*"),
                  values=c(mypal2)) +
scale_x_continuous(expand=c(0, 0)) +
labs(y=NULL,
     x="adjusted Rand index") +
theme_classic() +
theme(axis.text.x=element_markdown(),
      legend.text=element_markdown(),
      legend.position=c(0.9, 0.9),
      panel.grid.major.x=element_line(color="lightgray", size=0.25))

其中mypal2的定义为：

mypal <- colour("okabeito")(8)
mypal <- mypal[c(2:8, 1)]
names(mypal) <- NULL

mypal2 <- mypal[-c(2, 4, 6)]

palette(mypal2)

需要实现两个优化：

1. 将变量按**链接方法（single/complete/average）**分组展示
2. 把variable列中如pearson, single的标签，改为第一行显示相似性度量（如pearson）、第二行显示斜体链接方法（如*single*）的格式

要实现需求，需先处理数据结构，再修改绘图代码，具体步骤如下：

1. 数据预处理 #

拆分variable列，生成带格式的新标签，并按链接方法排序：

用data.table语法处理： #

library(data.table)

# 确保数据为data.table格式
setDT(b2)

# 拆分variable列为度量和链接方法两列
b2[, c("metric", "linkage") := tstrsplit(variable, ", ", fixed=TRUE)]

# 生成带换行和斜体的新标签
b2[, new_variable := paste0(metric, "<br>*", linkage, "*")]

# 按链接方法排序，保证同组变量相邻
b2[, linkage := factor(linkage, levels=c("single", "complete", "average"))]
b2 <- b2[order(linkage, metric)]
# 将新标签转为因子，锁定排序顺序
b2[, new_variable := factor(new_variable, levels=unique(new_variable))]

用tidyverse语法处理： #

b2 <- b2 %>%
  separate(variable, into=c("metric", "linkage"), sep=", ", remove=FALSE) %>%
  mutate(
    linkage = factor(linkage, levels=c("single", "complete", "average")),
    new_variable = paste0(metric, "<br>*", linkage, "*")
  ) %>%
  arrange(linkage, metric) %>%
  mutate(new_variable = factor(new_variable, levels=unique(new_variable)))

2. 修改绘图代码 #

替换y轴变量为新标签，启用Markdown解析，同时添加分组面板：

ggplot(b2, aes(y=new_variable, x=ARI, fill=dataset)) + 
  geom_col(position=position_dodge(), width=0.6) +
  # 按链接方法纵向分组，每组独立显示y轴
  facet_grid(linkage ~ ., scales="free_y", space="free_y") +
  scale_fill_manual(name=NULL,
                    breaks=c("pcaZ0", "pcaZ1", "modpcaZ0", "modpcaZ1"),
                    labels=c("非标准化", 
                             "标准化", 
                             "非标准化<br>*不含71-76*",
                             "标准化<br>*不含71-76*"),
                    values=mypal2) +
  scale_x_continuous(expand=c(0, 0)) +
  labs(y=NULL,
       x="调整兰德指数") +
  theme_classic() +
  theme(
    axis.text.y=element_markdown(),  # 解析y轴的Markdown格式标签
    legend.text=element_markdown(),
    legend.position=c(0.9, 0.9),
    panel.grid.major.x=element_line(color="lightgray", size=0.25),
    strip.background=element_blank(),  # 去掉分组面板的灰色背景
    strip.text.y=element_text(angle=0, hjust=0)  # 调整分组标签的显示角度
  )

可选调整： #

如果不需要分面板展示，仅需同链接方法的变量相邻，去掉facet_grid相关代码即可，此时new_variable的因子顺序已保证分组效果。

内容的提问来源于stack exchange，提问作者wantingtoimprove