ggplot2中仅修改分类变量单个层级的图例形状

问题根源是你将colour = peaks放在全局美学映射中时，geom_pointrange()会默认对所有分组应用「点+误差线」的几何样式，哪怕eprint_peaks组的误差值全为NA，图例也会统一沿用pointrange的默认样式。
你可以用两种方式实现目标效果： #

方法1：拆分图层（推荐，稳定性最高） #

将两个分组拆分为独立的绘图图层，eprint_peaks仅用geom_point()绘制圆点，mean_random_peaks用geom_pointrange()绘制点加误差线，两个图层共用colour映射即可自动生成正确的图例：

library(ggplot2)
library(data.table)

ggplot(data = dat, aes(x = factor(overlaps))) +
  # 绘制eprint_peaks层，仅圆点
  geom_point(
    data = ~.x[peaks == "eprint_peaks"],
    aes(y = cm_pct, colour = peaks),
    size = 2
  ) +
  # 绘制mean_random_peaks层，点+误差线
  geom_pointrange(
    data = ~.x[peaks == "mean_random_peaks"],
    aes(y = cm_pct, ymin = cm_pct - pct_sd - 0.5, ymax = cm_pct + pct_sd + 0.5, colour = peaks),
    size = 0.8, fatten = 2.5 # fatten控制点的相对大小，和上层圆点尺寸匹配
  ) +
  scale_x_discrete(
    name = "overlaps",
    breaks = seq_along(1:12),
    labels = paste0('\u2265', seq_along(1:12))
  ) +
  theme_classic(base_size = 13) +
  scale_colour_grey(
    labels = c("eprint_peaks" = "eprint\npeaks", "mean_random_peaks" = "random\npeaks")
  ) +
  labs(y = 'cumulative percentage', colour = NULL) + # 注意这里映射的是colour不是fill
  theme(
    legend.position = "top",
    legend.key.size = unit(1, 'cm')
  )

这种方法不需要手动调整图例参数，只要图层的美学映射一致，ggplot2会自动合并图例，同时你可以单独调整两个图层的点大小、线条粗细等参数，不会互相影响。 #

方法2：手动覆盖图例属性 #

如果不想拆分图层，可以在guides()中通过override.aes给不同图例项单独设置样式，注意参数向量的顺序必须和peaks因子的水平顺序完全一致：

ggplot(mapping = aes(x=factor(overlaps),y=cm_pct,colour=peaks),data = dat) +
  geom_pointrange(aes(ymin=cm_pct-pct_sd-.5,ymax=cm_pct+pct_sd+.5), fatten = 2.5) +
  scale_x_discrete(name="overlaps",breaks=seq_along(1:12),labels=paste0('\u2265',seq_along(1:12))) +
  theme_classic(base_size = 13) + 
  scale_colour_grey(labels=c("eprint_peaks"="eprint\npeaks","mean_random_peaks"="random\npeaks")) +
  labs(y='cumulative percentage',colour=NULL) +
  theme(legend.position = "top", legend.key.size = unit(1,'cm')) +
  guides(colour = guide_legend(
    override.aes = list(
      linetype = c(0, 1), # 第一个分组（eprint）隐藏误差线，第二个显示
      size = c(1.5, 0.8) # 匹配点和线条的尺寸
    )
  ))

注意：你原代码中labs(fill = NULL)是无效的，因为当前映射的美学是colour而非fill，需要改成colour = NULL才能正常隐藏图例标题。

内容的提问来源于stack exchange，提问作者moth