Interpreting DEseq2 output

Question

[roy.granit]

I have been using DEseq2 for quite a while, but today I was looking a certain transcript and compared that to the raw counts, and found little login in the results. These are the output values:

log2FoldChange: 0.3434941483
pvalue: 2.16740140052804e-06
padj: 5.65192536063831e-06

These are the raw counts:

Treament A: 1878,644,3010  
Treament B: 663,489,457

After RPM normalization I get these values:

Treament A: 993,996,1003 
Treament B: 977,1001,961

With these values, one can hardly notice any change.. Could it be that DEseq2 is looking at the data in such a different way so the RPM values make little sense?

Thanks!

[caggtaagtat]

Do you get the same results, when using TPM insteed of RPM?

[roy.granit]

The lib is 3' based, so I believe that the transcript length is not relevant in this case

Answer 1

[rpolicastro]

DESeq2 normalizes the reads for differences in library size using the geometric mean of counts. You can access the normalized counts using counts(dds, normalized=TRUE) after size factor estimation.

Regardless of method, the magnitude of the effect size from your example data is pretty low. For most downstream analysis, people choose a Log2 fold cutoff value on top of an adjust p-value threshold to only retain genes with a plausibly interesting effect size. This is also because p-value is partially related to the sample size due to the concept of statistical power and type II error (false negatives). The more samples you add, the better your ability to potentially detect smaller effect sizes in your sample, but whether these small effect sizes detected is interesting is an informed judgment call.

[roy.granit]

Thanks, using counts(dds, normalized=TRUE) I was indeed able to obtain the normalized counts and obtain a better understanding of the data..