How to normalize my results after DESeq2

Question

[nvijay1991]

Hi All,

I have a set of five strains effecting a crop and the corresponding RNA-Seq data and HTSeq-Counts from which I have generated DEGs up and down regulated in the crop when infected with each of this five strains.

Now I'm getting highly up and down regulated genes in few strains and low number of up and down regulated in few strains, I want to normalize this values to avoid biased interpretation with respective any particular strain.

Are there any literature/tools available to normalize the values in this situation and to get some uniform values across this five strains..!!

For your information, I have normalized and removed batch effects prior to calculating DEGs.

Thanks in advance.

Regards, Vijay Narsapuram

[Kevin Blighe]

How did you remove batch effects, and what evidence did / do you have that batch effects exist? Frequently, I come across researchers who worry too much about batch effects. Attempting to adjust for a batch effect where none exists can mess up your data.

[nvijay1991]

I had used Combat for checking if there are any presence of batch effects and see that there is no need to perform any batch correction. so I had supplied the raw HTSeq-Counts to DESeq2 tool the one in Galaxy.

[Kevin Blighe]

Okay, so, you did not correct for batch. How does your PCA bi-plot appear? How does the dispersion plot appear? In your own mind, the issue is that there is an imbalance in the number of statistically significantly differentially expressed genes in the comparisons that you have performed, right? How many samples do you have in each group?

[GenoMax]

Did you follow standard DESeq2 analysis protocol? You can access normalized counts from DESeq2 by: https://support.bioconductor.org/p/66067/

[nvijay1991]

I followed DESeq2 under Galaxy online tool to generate my DEGs

Answer 1

[swbarnes2]

For your information, I have normalized and removed batch effects prior to calculating DEGs.

You put the DESeq2 tag, but for DESeq, you aren't supposed to do any of that yourself. You are supposed to give DESeq raw gene counts. And DESeq doesn't "remove batch effects"; it includes batch in the linear model when 'batch' is part of the design.

[nvijay1991]

Sorry for the miss communication, I had supplied raw HTSeq-Counts to DESeq2. I was simultaneously checking for the batch effects presence using other tools like DEBrowser. As I mentioned now, I'm looking for tools to perform the normalization on the DEGs generated by DESeq2. One tool suggestion was to use metaP, but I'm still very skeptical about it.

[swbarnes2]

You want to normalize the genes? I'm not sure why you would expect to have uniform values across all your strains.

Answer 2

[ariel]

As @swbarnes2 said, it sounds very much like you are using the right tool for the wrong job.

DESeq isn't so much a package as an implementation of an algorithm for performing DE on raw RNA-Seq counts, where an important step in the algorithm is normalizing the counts. This process uses a special algorithm and a specific probabilistic model that has been validated for RNA-Seq data. There are other steps in there as well (Bayesian priors, effect size shrinkage, etc.).

Either that's the algorithm you want, or you want a different one.

If you start messing with your data before putting it into DESeq then no one can really tell you what you are getting out.

Answer 3

[nvijay1991]

Issue has been solved and I could figure out the mistake and resolved it, Thanks for the timely reply's and comments.