My PCA plot looks good but i don't have differentially expressed genes

Question

[druggable]

Dear All,

I have the following PCA plot from microarray data:

It looks quite good because the green samples and red samples, which i am interested in, are well separated along PC1. However, when i ran differential expression analysis, there are no significant differentially expressed genes and the histogram looks like this:

Is this result normal, or to be expected?

Thanks!

[Papyrus]

You should use the raw p-values, not the adjusted p-values, to inspect the histogram

Answer 1

[Gordon Smyth]

The PCA plot shows what seems to be a massive batch effect (left to right) and also one outlier (purple point at the bottom). The green vs red group comparison is confounded with the batch effect, so getting results will be difficult.

Can't make any comments about your DE results without knowing how you analysed the data. There are ways to account for batch effects and outliers.

Answer 2

[Jautis]

Your sample size is small, which may be one reason you don't detect a lot of differential expression. That being said, I would still expect to see some signal, so it may come down to how you're processing your data? It's almost impossible to tell from the information given.

Gordon's point about batch effects is also a good one -- what does each color represent in the PCA?

[druggable]

The colors represent phenotypes: healthy (pink), obese + severe diabetes (green), obese + well controled diabetes (blue), and obese only (purple). Unfortunately, there is no information on the batch effect.

I started with raw cel files and did the following:

data = ReadAffy(celfile.path=celpath)

data.rma.norm = affy::rma(data)

rma = exprs(data.rma.norm)

Then I used prcomp for the PCA.