Heatmap correlation with pvalues

[mslider]

Hi,

i'm looking for code or ideas to draw correlation heatmap (spearman) with pvalue significances (*, **, ***), such as the picture i send here.

thank you --

[AlexisW]

There are 3 questions in here:

1. how to get the Spearman correlation coefficients, you can use cor(x, y, method = "spearman")
2. how to compute the p-values, that will depend on your context, the most natural one is probably cor.test(x, y, method = "spearman")
3. how to how to draw a heatmap with the stars, here you can use pheatmap() in the eponymous package, and use display_numbers = matrix where you precompute a matrix with the appropriate number of stars.

Of course since you have a lot of computations to make, you might need to use a for loop or similar to do all these, the easiest approach depends on what your data looks like.

[AlexisW]

Here is an example code (generic code though, you may need to change the test etc to suit your situation):

library(pheatmap)
set.seed(123)

# create some fake data
nb_samples <- 50
metabolites <- data.frame(LysoPC = runif(nb_samples, 0, 100),
                          PE = runif(nb_samples, 0, 100))

echocardiographic_parameters <- data.frame(LVEDD = 5 * metabolites$LysoPC + rnorm(nb_samples,sd = .01),
                                           LVEF = 2 * metabolites$PE - 3 * metabolites$LysoPC + rnorm(nb_samples, sd = .1))

# initialize empty matrices
mat_cor <- mat_pval <- mat_signif <- matrix(NA,
                                            nrow = ncol(metabolites),
                                            ncol = ncol(echocardiographic_parameters),
                                            dimnames = list(colnames(metabolites),
                                                            colnames(echocardiographic_parameters)))

# for each pair, compute one cell of the matrice
for(metab in colnames(metabolites)){
  for(param in colnames(echocardiographic_parameters)){
    
    # correlation coefficient
    mat_cor[metab, param] <- cor(metabolites[[metab]],
                                 echocardiographic_parameters[[param]],
                                 method = "spearman")
    
    # pvalue
    pval <- cor.test(metabolites[[metab]],
                     echocardiographic_parameters[[param]],
                     method = "spearman")[["p.value"]]
    
    # decide nb of stars
    if(pval >= 0.05){
      stars <- ""
    } else if(pval >= 0.01){
      stars <- "*"
    } else if(pval >= 1e-3){
      stars <- "**"
    } else{
      stars <- "***"
    }
    
    # save pvalue and stars in the corresponding matrices
    mat_pval[metab, param] <- pval
    mat_signif[metab, param] <- stars
  }
}


# check that we could have directly computed all the coefficients without a loop
all.equal(
  mat_cor,
  cor(metabolites,
      echocardiographic_parameters,
      method = "spearman")
)
#> [1] TRUE



pheatmap(mat_cor,
         display_numbers = mat_signif,
         fontsize_number = 20)

^{Created on 2023-11-06 with reprex v2.0.2}

[mslider]

Hi, sorry for the delay.

yes, congratulation,
this is what i want to obtain, I'll try to reproduce your method using matrices like you.
thank you so much.