CQN to DESeq2
1
0
Entering edit mode
bioinfo • 0
@bioinfo-12782
Last seen 6 weeks ago
United States

Hi, I am analyzing RNA-seq and want to put cqn offset (GC% normalization) to DESeq2 object. I followed the vignette of DESeq2 but have a question about geometric mean standardization of cqn offset.

normFactors <- exp(cqn.out\$glm.offset)
summary(rowMeans(normFactors))


Min., 1st Qu., Median, Mean, 3rd Qu., Max.
0.05063, 0.47634, 0.63402, 0.76283, 1.11039, 1.45960

normFactors1 <- normFactors / exp(rowMeans(log(normFactors)))
summary(rowMeans(normFactors1))


Min., 1st Qu., Median, Mean, 3rd Qu., Max.
1.009, 1.822, 2.309, 2.437, 3.022, 6.235.

I saw that the purpose of geometric mean standardization is that the mean of normalized counts for a gene is close to the mean of the unnormalized counts. But in this case, without the standardization, the distribution of normFactors are close to 1. Could you give any suggestion? Do I need a standardization in this case?

Also, could you explain the motivation of geometric mean standardization more specifically? why do we nee to make the mean of normalized counts for a gene close to the mean of the unnormalized counts ?

cqn DESeq2 • 225 views
1
Entering edit mode
@mikelove
Last seen 3 hours ago
United States

I'd just follow the code -- note that it only changes the intercept.

You want the two means to be similar because the mean of normalized counts is used as a proxy for E(y) in various places in the method, including dispersion trend.