Make rectangular matrix of correlation values in R, possibly using corrplot

Question

I want to make a matrix of correlation values in R. However, I am not comparing the parameters to themselves, so it isn't your typical corrplot that is a triangle and mirrored over a diagonal axis. I actually already have the correlation values.

I was just hoping to recruit the corrplot function (from corrplot package), which produces images like the following:

My data, already containing the correlation values I want to plot, is:

            X    animal.1   animal.2     animal.3    animal.4   animal.5
1 parameter 1  0.10258087  0.3338782  0.150246554  0.07295133  0.2484152
2 parameter 2  0.04205748  0.4062727 -0.002101464  0.12068818  0.2951127
3 parameter 3  0.11264488  0.4114954  0.067145776  0.13361071  0.3246052
4 parameter 4 -0.02261649 -0.2426341  0.108042167 -0.12820517 -0.2005686
5 parameter 5 -0.01576384 -0.2300852  0.112941655 -0.12391976 -0.1906473
6 parameter 6 -0.09749030 -0.3110920  0.021994297 -0.13570257 -0.2557532

I want the plot to look like a corrplot, but it will be a full rectangle. I want the circles of varying diameters, and the two different colors for positive or negative values, and the color gradient along the side of the plot. In essence I want the corrplot theme of a rectangular table of values.

I have also asked to see if asterisks could be placed over the grid squares with significant (p<0.05) p-values. I am now attaching a dput file of the p-value matrix

structure(list(X = structure(1:6, .Label = c("parameter 1", "parameter 2", 
"parameter 3", "parameter 4", "parameter 5", "parameter 6"), class = "factor"), 
Animal.1 = c(0.2454906, 0.63471, 0.2019519, 0.7984066, 0.8587147, 
0.2698293), Animal.2 = c(0.000103586, 1.62e-06, 1.15e-06, 
0.005412082, 0.008451775, 0.000315107), Animal.3 = c(0.08796553, 
0.9810688, 0.447827, 0.2211191, 0.2007585, 0.8038392), Animal.4 = c(0.4094627, 
0.1713835, 0.1296492, 0.1460429, 0.1601174, 0.1236984), Animal.5 = c(0.004374306, 
0.000653099, 0.00016464, 0.02213469, 0.0298038, 0.003315349
)), .Names = c("X", "animal.1", "animal.2", "animal.3", "animal.4", 
"animal.5"), class = "data.frame", row.names = c(NA, -6L))

Please do not provide your data as an image. No one wants to type it all in again. Instead, use dput to create a text version of your data and paste the result into your question. Also, please provide the code that you used to create the corrplot that you have now You probably should read How do I ask a good question? — G5W
– G5W, Commented Mar 14, 2020 at 0:59
Both pictures were meant to demonstrate what I wanted. I took a picture of a random corrplot picture example I found online. My data was provided merely to demonstrate the shape of the plot I wanted. I don't know what dput is. I'm new to stackoverflow — MapDeath
– MapDeath, Commented Mar 14, 2020 at 1:05
You must have got the correlation values from some data - why not just use the original data, which is what corrplot is designed for? — Richard Ambler
– Richard Ambler, Commented Mar 14, 2020 at 2:16
The original data is not formatted correctly, or otherwise I don't know how to extract bits of data. I am not that R savvy. I do everything clunkily but formulaically — MapDeath
– MapDeath, Commented Mar 14, 2020 at 3:20

Jonathan V. Solórzano · Accepted Answer · 2020-03-14 01:33:09Z

5

I think you just need to transform your df into a matrix and then run the corrplot function.

library(corrplot)

#Transform data to matrix
matrix_cor<-as.matrix(df[,-1])

#Set row names as df first column
row.names(matrix_cor)<-df[,1]

corrplot(matrix_cor,
         method = "circle")

answered Mar 14, 2020 at 1:33

Jonathan V. Solórzano

5,78516 silver badges26 bronze badges

Sign up to request clarification or add additional context in comments.

6 Comments

MapDeath Over a year ago

You did exactly what I was looking for. Upvote times a million, thanks man! Actually one more thing, is there a way to draw an asterisk over each of the circles, or indicate some other way on the plot, if it is a significant (p<0.05) correlation?

MapDeath Over a year ago

I extracted a p-value for animal 3 parameter 3 from cor.test and it is not significant like your code shows by putting an asterisk there

Jonathan V. Solórzano Over a year ago

Yeah, sorry, you have to run cor.mtest on the original data, not the correlation matrix. So, you can do result<-cor.mtest("original_data", conf.level = 0.95) and then

corrplot(matrix_cor,           p.mat = result$p,           insig = "label_sig",          sig.level = 0.05,           pch.cex = 1.5,           pch.col = "black")

MapDeath Over a year ago

I just made a new csv matrix by individually copy pasting all the p-values from cor.test. Can you use that instead and overlay it on the plot? I have added that table into the original question

MapDeath Over a year ago

Also, my original data is a different structure than these simplified datasets I am posting, so I can't reference the original data unfortunately

|

dc37 · Accepted Answer · 2020-03-14 03:30:56Z

As you tag this question with ggplot2, I'm providing you a way to do that using ggplot2.

First, you need to reshape both the dataframe containing correlation values (named df here) and the one with pvalues (named df_pval here). For doing that, you can for example use the pivot_longer function from tidyr package (also part of the tidyverse package).

On the dataframe containing pvalues, we will add a column Label in order to mark with an asterisk all values below 0.05.

library(tidyr)
library(dplyr)
library(ggplot2)

corr_values <- df %>% pivot_longer(-X, names_to = "Animal", values_to = "value") 

# Here an extract of the dataframe with correlation values after reshaping it:
  X           Animal    value
  <fct>       <chr>     <dbl>
1 parameter 1 animal.1 0.103 
2 parameter 1 animal.2 0.334 
3 parameter 1 animal.3 0.150 
4 parameter 1 animal.4 0.0730
5 parameter 1 animal.5 0.248 
6 parameter 2 animal.1 0.0421

corr_pval <- df_pval %>% 
  pivot_longer(-X, names_to = "Animal", values_to = "p.value") %>%
  mutate(Label = ifelse(p.value < 0.05, "*",NA)) 

# Here an extract of the dataframe obtained
  X           Animal    p.value Label
  <fct>       <chr>       <dbl> <chr>
1 parameter 1 animal.1 0.245    NA   
2 parameter 1 animal.2 0.000104 *    
3 parameter 1 animal.3 0.0880   NA   
4 parameter 1 animal.4 0.409    NA   
5 parameter 1 animal.5 0.00437  *    
6 parameter 2 animal.1 0.635    NA

Then, you can use geom_point and geom_tile to get the heatmap. You can also customize, color, size, etc.. of your plot by passing various functions such as scale_size_continuous and scale_color_gradient.

Finally, you can add asterisk for pvalues by calling a new dataframe corr_pval into geom_text and specify the appropriate aesthetic argument:

library(ggplot2)

ggplot(corr_values, aes(x = X, y = Animal))+
  geom_tile(color = "black", fill = "white")+
  geom_point(aes(color = value, size = abs(value)))+
  scale_color_gradient2(low = "green", mid = "white", midpoint = 0, high = "red", name = "")+
  scale_size_continuous(range = c(5,15), name = "")+
  geom_text(data = corr_pval, aes(label = Label), size = 8, vjust = 0.7, hjust = 0.5)

So, you should get something like that:

Collectives™ on Stack Overflow

Make rectangular matrix of correlation values in R, possibly using corrplot

2 Answers 2

6 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

6 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related