CODEWITHSUNDEEP
rfor-loopcorrelationsapply
keenan
keenan

Reputation: 504

R find correlations between dataframes of different sizes

I have two data frames with the same number of columns but different numbers of rows. I am trying to run correlation tests for all of the rows. I am able to do this with a for loop but because of the size of the dataset it is not a feasible option. I have found solutions for finding correlations of equal size data frames but I am not sure how to adapt these solutions.

Here is my for loop solution that works for smaller datasets.

c.mg.spearmanB = data.frame()
for (i in 1:nrow(brainMicroRNAs)) {
  for (j in 1:nrow(brainGenes)) {
    miRNA = brainMicroRNAs[i,]
    gene = brainGenes[j,]
    #calculate correlations and add to dataframe
    c.mg.spearmanB[i,j] = cor.test(miRNA, gene, method="spearman", exact=F)$p.value
  }
}

Upvotes: 0

Views: 931

Answers (2)

Axeman
Axeman

Reputation: 35392

Many R functions expect observations to be rows, and variables to be columns. You appear to have flipped this around. No worries, we can transpose and use the normal cor function.

Some example data:

set.seed(1234)
df1 <- as.data.frame(matrix(rnorm(12), ncol = 3))
df2 <- as.data.frame(matrix(rnorm(15), ncol = 3))

Now calculate the correlations:

cors <- t(cor(t(df1), t(df2)))

If you need p-values, we can do that manually, using vectorized functions:

df <- ncol(df1) - 2
t_vals <- cors * sqrt(df) / sqrt(1 - cors ^ 2)
p_vals <- 2 * min(pt(t_vals, df), pt(t_vals, df, lower.tail = FALSE))

Those p-values are two-sided.

And to check, if it all worked:

cor.test(unlist(df1[3, ]), unlist(df2[3, ]))
# Pearson's product-moment correlation
# 
# data:  unlist(df1[3, ]) and unlist(df2[3, ])
# t = -0.015874, df = 1, p-value = 0.9899
# alternative hypothesis: true correlation is not equal to 0
# sample estimates:
#         cor 
# -0.01587196 
cors[3, 3]
# [1] -0.01587196
p_vals[3, 3]
# [1] 0.9898952

Upvotes: 1

Raghu
Raghu

Reputation: 41

Using loops is not recommeded. Use df.corr of pandas instead. Merge the two dfs and then run the corr. If possible provide samples of datasets.

Upvotes: 0

Related Questions