which kind of normalization is suitable for corr.test

Question

[diqixiaoyaoer]

I decided to use corr.test to calculate the correlation between genes.

And I know the input object must be a matrix or dataframe. But I don't think it's a vital part for me to pay attention to.

Now what confuses me is I don't know which kind of normalized gene expression matrix is suitable for corr.test. I think the FPKM gene counts after t() is suitable before. But I don't think so after somebody told me something about vst-transformed counts ？

Can somebody give me some advice ？ Now I have the FPKM value and corr.test .If I need to change my normalization method ?

Vary thankful.

[diqixiaoyaoer]

I have several datasets and all of them were dealt with Galaxy and got the raw gene expression counts. And then，I merged them together. Now I want to know the correlation between genes. I know it can't be raw gene expression counts that to used in corr.test. And normalization is necessary. But I don't know which kind of normalization is suitable for corr.test. So can you give me some suitable advice ? Am I making myself clear ？

Answer 1

[basuanubhav]

I think using vst transformed counts will be suitable.

[i.sudbery]

or rLog. Both from the DESeq2 package.

Answer 2

[Rogerio Ribeiro]

With RNA-seq data, usually, there is a tendency for the variance to increase with expression level. I.e highly expressed genes usually have higher variance. You can easily visualize this by plotting your data (median expression per genes vs variance per gene) or checking this blog post I found. Due to this effect, to avoid any correlation-based test being "dominated" by highly expressed genes usually a variance stabilizing transformation is applied like the vst transformation from the DESeq2 package or a simple a log(x+1) transformation. I'm not sure if VST is meant to be applied to FPKM data, or if is just valid with raw counts. I suggest you try the log transformation and then plot your data to check for irregularities.

[diqixiaoyaoer]

Vary thankful !! I also search for related Information about my object. And I agree with what you said. I will have try of log transfomation. Thank you.