Correlating methylation and expression do i need normalization?

Question

[Pin.Bioinf]

Hello,

I need to correlate expression values with methylation values. Do I need to normalize expression values from my matrix? What R package do you recommend? Do I need to normalize the beta values for the CpGs?

Thank you in advance

[Pin.Bioinf]

Thank you very much,

I realized I have rpkms so it is normalized. I applied smoothing to the methylation values, and I am wondering if i do need to normalize that anyways?

I will take a look at the parallel implementation, it will help me a lot.

Thank you again

[Kevin Blighe]

De nada amigo. Nos vemos / Hasta luego.

PD - if you have beta values, then these are technically already normalised.

PD - if you do want to have the 2 datasets on the same distribution, then I suggest to convert them to the Z-scale with the scale() function in R

Answer 1

[Kevin Blighe]

Your starting point for what you want to perform is indeed a normalised dataset, for both the expression and methylation data. They do not necessarily have to be on the same scale but they should both be normalised. In addition, use Spearman correlation, not Pearson.

Then, provided that your samples are matched between both datasets, you should be able to correlate them easily. Be wary of the fact that, if you attempt to correlate something like 20,000 genes to 34,000 methylation probes, then you will crash R. In this light, take a look at bigcor: https://www.r-bloggers.com/bigcor-large-correlation-matrices-in-r/

Kevin