Question

How to make boxplot with pairwise and overall p value from Limma

0

Entering edit mode

Chris ▴ 20

@3fdb6f97

Last seen 3 months ago

United States

Hi all,

I use limma to compare pathway score between 3 groups. Then I would like to make boxplot with overall p value and pairwise p value. I use this with method ='t.test' or method='anova':

ggboxplot(data, x = "disease_state", y = "pathway_name",
          color = "dem8", palette = "jco") + 
  stat_compare_means(comparisons = my_comparisons, method = 't.test') + # Add pairwise comparisons p-value
  stat_compare_means(label.y = 0.6, method = 't.test') +
  labs(y = "Pathway Score", title = "pathway_name")

http://www.sthda.com/english/articles/24-ggpubr-publication-ready-plots/76-add-p-values-and-significance-levels-to-ggplots/

The value from t test or anova is different from the p value from limma:

topTable <- topTable(fit2, number=Inf, sort.by="none")
topTableLSvsNC <- topTable(fit2, coef="LSvsNC", number=Inf, sort.by="none")

Would you please have an explanation for the difference and how to make boxplot with overall p value and pairwise p value from limma? Is that because statistical model difference which limma is better? I use a manual way: extract p value from topTable and then use geom_text to annotate to boxplot. But for pairwise p value I don't know how. Thank you so much!

enter image description here

t.test boxplot limma anova • 2.0k views

ADD COMMENT • link 8 months ago Chris ▴ 20

1

Entering edit mode

Regarding the difference between the p-value from a t-test and a p-value from limma, this has been previously discussed in this forum, at least in this post.

ADD REPLY • link 8 months ago Robert Castelo ★ 3.4k

0

Entering edit mode

Thanks Robert. I think p value from limma is better, now just try to add them to boxplot.

ADD REPLY • link 8 months ago Chris ▴ 20

0

Entering edit mode

Read LIMMA: where to find the ANOVA?

topTable(fit2) table reports the ANOVA p-value.

topTableLSvsNC reports the p-value for the specified contrast. Loop over the contrasts to get the "pairwise" p-values. These p-values are adjusted per contrast.

ADD REPLY • link 8 months ago SamGG ▴ 360

0

Entering edit mode

Yes, I know how to extract these p value but how to add them to the boxplot in a nice way like the example plot above?

ADD REPLY • link 8 months ago Chris ▴ 20

1

Entering edit mode

ggpubr is based on ggsignif. Read the documentation of geom_signif.

All you need is the list of comparisons and the corresponding text to place on top of the horizontal bar.

Here is a simple example using that helps you building this.

library(ggplot2)
library(ggsignif)
ggplot(mpg, aes(class, hwy)) +
    geom_boxplot() +
    geom_signif(
        annotations = c("**", "***"),
        comparisons = list(c("2seater", "compact"), c("midsize", "minivan"))
    )

custom aanotations

ADD REPLY • link 8 months ago SamGG ▴ 360

0

Entering edit mode

Thank you so much! I try now, hope it works. It worked. I try to figure out how to make a boxplot and then try to combine 11 boxplots into one figure. For only one overall p value, this worked:

long_data <- merge(long_data, p_values_df, by = "Pathway")
ggplot(long_data, aes(x = dem8, y = Score, fill = dem8)) +
  geom_boxplot() +
  facet_wrap(~ Pathway, scales = "free_y", ncol = 4) +  # Adjust scales and ncol as needed
  labs(title = "", x = "Disease state", y = "Pathway Score") +
  theme_light() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1),  # Rotate x-axis text for readability
        strip.text.x = element_text(size = 10)) +  # Adjust size of facet labels if needed
  geom_text(aes(x = Inf, y = Inf, label = annotation_text),
            hjust = 1.1, vjust = 2, size = 4, colour = "red", inherit.aes = FALSE)

I used pivot_longer() to make long_data. This way, I don't have to manually annotate each plot. Do you have any suggestion for pairwise p value for multiple boxplot?

ADD REPLY • link 8 months ago Chris ▴ 20