Change shape of the data point based on positive or negative value in ggplot2

Question

I have the following dataset:

beta pval Category_name
0.5 0.005 One
-0.3 0.6 Two
0.2 0.03 Three
-0.1 0.7 Four

I am trying to make a figure using pval (y axis) and Category_name (x axis). I would like to use shape 24 when beta is positive, and shape 25 when beta is negative, but still plotting pval.

My script looks like this:

library(ggplot2)
library(RColorBrewer)

colourCount <- length(unique(dup_categories_merged_allSNP$Category_name))
getPalette <- colorRampPalette(brewer.pal(9, "Set1"))

dup_categories_merged_allSNP$shape <- ifelse(dup_categories_merged_allSNP$beta > 0, 24, 25)

PheWAS_all <- ggplot(dup_categories_merged_allSNP, aes(x = Category_name, y = -log10(pval), shape = shape, colour = Category_name)) + 
geom_point(size = 2) +
  geom_jitter() +
  theme_classic() + 
  scale_colour_manual(values = getPalette(colourCount)) +
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 0, size = 10),
        axis.title.x = element_blank(),
        legend.position = "none") + 
  labs(color = "Category_name", x = "", y = "-log10(pval)") +
  geom_hline(yintercept = -log10(3.65E-09), color = "darkmagenta", linetype = "dashed", size = 1, alpha = 0.5) +
  geom_hline(yintercept = -log10(0.000740457), color = "gray32", linetype = "dashed", size = 1, alpha = 0.5)

I get the following error message: Error in scale_f():! A continuous variable can not be mapped to shape.

Edit: Original figure:

Figure with the changes based on answer(linear on the y axis):

Subset of the data:

Change your ifelse to `ifelse(dup_categories_merged_allSNP$beta > 0, "high", "low")` and add `+scale_shape_manual(values = 24:25)` to your plot — Allan Cameron, Mar 23 '23 at 21:27
Thanks! This is great, but it doubles the points - instead of showing one data point once for pval, it shows another one for beta. Do you know how to avoid this? — kllrdr, Mar 24 '23 at 10:36
I guess you x-axis also has a continuous scale. Can you provide a subset of your data? May be a column is missing in your current small dataset here for x- axis? — S-SHAAF, Mar 24 '23 at 12:18
I added it to the question. I have not used anything but beta, pval, and category_name, but there is an additional column, "phenotype," that could be useful. Among the same Category_name, there are several different phenotypes. — kllrdr, Mar 24 '23 at 13:23
In your Manhattan plot the p-values are plotted against phenotype values within each category, right? — S-SHAAF, Mar 24 '23 at 15:29
Actually I did not specify phenotype anywhere, even on the first plot. The p values are plotted against Category_name. So this is why I don't really understand what is happening in the new plot. — kllrdr, Mar 24 '23 at 17:34
In the first figure in the answer, there are two triangles. I think the small one is the one I want to see on the final figure, those seem to be dispersed and not linear to the y axis. Do you know why this happens and how to solve it? — kllrdr, Mar 24 '23 at 18:04
OK, To produce better plot I generated a random dataset because your sample data are too small (only few rows). see my updated answer — S-SHAAF, Mar 24 '23 at 18:51

S-SHAAF · Accepted Answer · 2023-03-24T18:46:28.623

One way is to create a column using mutate() and ifelse() condition using tidyverse. To show the plot I generated example dataset to increase size of data. For the shapes, you can add scale_shape_manual() in your ggplot function and your favorite shape values inside the function. I added also legend to indicate the shapes but you could remove it from the code:

some example data

library(tidyverse)
library(ggplot2)
library(RColorBrewer)
library(ggh4x)

set.seed(1)
phenotype <- sample(1:3, 80, replace=TRUE)
beta <- runif(80, min=-1, max=1)
pval <- runif(80, min=0.005, max=1)
Category_name <- rep(c("One", "Two", "Three","Four"), each=20)
dup_categories_merged_allSNP <- data.frame(phenotype,beta,pval,Category_name)

add shape category based on beta values and provide data for plot

dup_categories_merged_allSNP <- mutate(dup_categories_merged_allSNP,shape = ifelse((beta < 0), "negative", "positive"))

dup_categories_merged_allSNP$Category_name <- factor(dup_categories_merged_allSNP$Category_name
                                                     ,levels = c("One","Two","Three","Four"))

colourCount <- length(unique(dup_categories_merged_allSNP$Category_name))
getPalette <- colorRampPalette(brewer.pal(9, "Set1"))

Create plot with shapes based on positive/negative values. The phenotype values must be added to avoid overlapping pvalues within each category so:

PheWAS_all <- ggplot(dup_categories_merged_allSNP, aes(x = factor(phenotype), y = -log10(pval))) + 
  
  geom_point(aes(x=factor(phenotype),shape=shape, color=Category_name, fill=Category_name),size = 2.5)+
  theme_classic() + 
  scale_color_manual(values = getPalette(colourCount)) +
  scale_fill_manual(values = getPalette(colourCount)) +
  scale_shape_manual(values = c(25, 24)) +
  facet_grid2(. ~ Category_name, scale = "free_y",switch = "x", strip = strip_themed(background_x = elem_list_rect(fill=getPalette(colourCount))))+
  scale_y_continuous(expand = c(.01, 0) ) + 
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 0, size = 10),
        axis.title.x = element_blank(),
        legend.position = "none")+
  theme(legend.position = "right")+
  theme(axis.text.x = element_blank()) +
  theme(axis.ticks = element_blank()) +
  theme(panel.spacing = unit(0, "lines")) +
  labs(x=NULL) +
  labs(color = "Category_name", x = "", y = "-log10(pval)") +
  geom_hline(yintercept = -log10(3.65E-09), color = "darkmagenta", linetype = "dashed", size = 1, alpha = 0.5) +
  geom_hline(yintercept = -log10(0.000740457), color = "gray32", linetype = "dashed", size = 1, alpha = 0.5)+
theme(strip.text = element_text(vjust =.8,face = "bold",
                                size = 9,margin = margin(b = 3)))
PheWAS_all

Thank you so much! The only thing I would like to change is having double points for each data point - I want to only plot pval, and the orientation of the shape based on beta. For one, three, and four, there are double points on the figure. — kllrdr, Mar 24 '23 at 10:41
Ok, I added this to my answer showing only pvalues. please see my updated answer — S-SHAAF, Mar 24 '23 at 11:12
Thank you! One last thing - the dispersion of the values on the figure are linear on the y axis. Is their any way to keep the original dispersion of the p values? I added the figure to my question to show what I mean. — kllrdr, Mar 24 '23 at 11:44

Change shape of the data point based on positive or negative value in ggplot2

1 Answers1