0
$\begingroup$

I am working on a script to create a correlation matrix and mutual_rank matrix from RNAseq gene counts for 57992 genes in 7027 samples. I have already tested this script for a smaller file (for 10,000 genes) and it is working, however, when I am running this script on my final input file, it is giving the following error:

 corrupt matrix -- dims not not match length (this error comes after gene to 
 gene correlation step)

Script which i am running is:

E <- read.table("mdat_human_test.txt", header = TRUE, row.names=1)

# gene-to-gene correlation
C <- cor (t(E))
write.table(C, "human_correlation_genes_mdat.txt", quote=FALSE, sep="\t", 
row.names=TRUE)

# correlation rank matrix
R <- t(apply(1-C, 1, rank))
write.table(R, "human_rank_genes_mdat.txt", quote=FALSE, sep="\t", 
row.names=TRUE)

# mutual rank matrix
M <- sqrt(R * t(R))
write.table(M, "human_mutual_rank_genes_mdat.txt", quote=FALSE, sep="\t", 
row.names=TRUE)

q()
$\endgroup$
1
  • 1
    $\begingroup$ What is your biological question of interest ? Why are you doing this correlations? In addition to Devon's answer, I would like to point out that if you are using all the human annotated genes you might need to filter and normalize them. You'll presumably will have 25% of "genes" which are poorly described and/or with few counts are you really interested in keeping those ? $\endgroup$
    – llrs
    Mar 6 '19 at 13:40
3
$\begingroup$

You can't compute a correlation matrix when you have more than ~46000 rows, since a standard R matrix can have a maximum of 2^31-1 values. Have a look at packages like bigcor. Alternatively, consider if having 3+ billion correlation values is really useful.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.