Raw Reads To Calculate Differential Gene Exrepssion

Question

[camelbbs]

Hi all,

I am now using raw reads to calculate the differential gene expression. I see DESeq gives SizeFactors for each sample and can use that for normalization. But it seems DESeq only use that normalization for visualization, not for calculate the differential pvalue. Is that correct?

I've read DESeq paper and still not very clear about that. I know DESeq use NB distribution to calculate the pvalue. But for raw reads, it must be normalized first, right?? Thanks,

Che

Answer 1

[Steve Lianoglou]

I'm curious what makes you think that the sizeFactors are only used for visualization?

If you download the source code to for the package and poke around a bit, you'll see that the sizeFactors are used everywhere from dispersion estimation, nbinomTest-ing, fitting GLMs, getting variance stabilized data, etc.

[camelbbs]

Thanks a lot Steve. I may make a mistake. Is the sizeFactors the only normalization method used in DESeq? After the sizeFactors normalization, can the value be used for fold-change detection between two samples?

[Steve Lianoglou]

Have you read the DESeq vignette? Fold changes are calculated for you after a call to nbinomTest, for example. To just get something like a matrix of expression that you can calculate fold changes for data exploration, look at the example in section 6, where they shoot their data through a variance stabilizing transformation.

Answer 2

[camelbbs]

As my thinking, sizeFactors is a similar method to RPM normaliztion, the only difference is RPM use all reads in one sample as "sizeFactors", but sizeFactors use the median scaled counts. Can anyone confirm this?