如何在ggplot分簇柱状图展示lme模型lsmeans()校正配对比较p值？

Great question! When working with lme models and lsmeans post-hoc comparisons, adding significance markers to grouped/clustered ggplot bar plots is totally manageable with a few specialized packages. Let’s walk through both the quick-and-easy method for asterisks/connecting lines, plus some more elegant alternatives for presenting all pairwise results.

Quick Method: Asterisks & Connecting Lines #

The ggpubr package is my go-to for this—it’s designed to work seamlessly with ggplot2 and handles clustered groups smoothly. Here’s a step-by-step breakdown:

1. Extract and format your lsmeans results
   First, pull the pairwise comparisons from your lsmeans output and clean them up to include significance labels (e.g., * for p<0.05, ** for p<0.01, etc.):

   library(lsmeans)
   library(dplyr)
   library(stringr)
   
   # Replace with your model and grouping variables
   lsmeans_comps <- lsmeans(your_lme_model, pairwise ~ treatment | cluster)
   comps_df <- as.data.frame(lsmeans_comps$contrasts)
   
   # Add significance labels
   comps_df <- comps_df %>%
     mutate(
       sig = case_when(
         p.value < 0.001 ~ "***",
         p.value < 0.01 ~ "**",
         p.value < 0.05 ~ "*",
         TRUE ~ "ns"
       ),
       # Split contrast into individual groups for plotting
       group1 = str_split_fixed(contrast, " - ", 2)[,1],
       group2 = str_split_fixed(contrast, " - ", 2)[,2]
     )
   

2. Add markers to your ggplot
   Use stat_pvalue_manual from ggpubr to overlay the significance lines and asterisks. Make sure the position_dodge width matches your bar plot to keep alignment correct:

   library(ggplot2)
   library(ggpubr)
   
   # Replace with your plot data and aesthetics
   clustered_bar_plot <- ggplot(your_plot_data, aes(x = treatment, y = response, fill = cluster)) +
     geom_col(position = position_dodge(width = 0.8), width = 0.7) +
     # Add significance markers
     stat_pvalue_manual(
       comps_df,
       x = "treatment",
       xmin = "group1",
       xmax = "group2",
       label = "sig",
       position = position_dodge(width = 0.8),
       tip.length = 0.01,
       size = 4
     ) +
     theme_bw()
   
   print(clustered_bar_plot)
   

More Elegant Alternatives for Many Comparisons #

If you have a lot of pairwise comparisons, cramming all those lines/asterisks into your bar plot can get cluttered. Here are two cleaner approaches:

1. Significance Heatmap #

A heatmap lets you visualize all pairwise p-values at a glance, with facets for your clusters. It’s perfect when you have multiple groups to compare:

# Reshape data for heatmap
heatmap_data <- comps_df %>%
  mutate(
    # Create unique pair labels (avoid duplicates like A vs B and B vs A)
    pair = paste(pmin(group1, group2), pmax(group1, group2), sep = " vs ")
  ) %>%
  distinct(cluster, pair, p.value, sig) %>%
  separate(pair, into = c("group_x", "group_y"), sep = " vs ")

# Build the heatmap
sig_heatmap <- ggplot(heatmap_data, aes(x = group_x, y = group_y, fill = p.value)) +
  geom_tile(color = "white") +
  geom_text(aes(label = sig), color = "black", size = 4) +
  scale_fill_gradient(low = "#90EE90", high = "#FF6347", limits = c(0, 0.05)) +
  facet_wrap(~cluster) +
  theme_bw() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

print(sig_heatmap)

2. Plot + Inline Comparison Table #

For full transparency, pair your bar plot with a concise table of all post-hoc results using gridExtra. This keeps your plot clean while giving readers access to exact p-values:

library(gridExtra)

# Format the comparison table
comp_table <- comps_df %>%
  mutate(p.value = round(p.value, 3)) %>%
  select(Cluster = cluster, Comparison = contrast, `P-Value` = p.value, Significance = sig)

# Convert table to a plot-friendly grob
table_grob <- tableGrob(comp_table, rows = NULL, theme = ttheme_minimal())

# Combine plot and table
grid.arrange(clustered_bar_plot, table_grob, ncol = 2, widths = c(2, 1))

Final Tips #

• If you only have a handful of comparisons, the asterisk/line method is intuitive and reader-friendly.
• For large numbers of comparisons, the heatmap or plot+table combo avoids visual overload and makes results easier to parse.

内容的提问来源于stack exchange，提问作者Alex