Entering edit mode
9.7 years ago
pmanga
▴
60
Hi,
I have this normalized dataset that I want to analyze the same way as DESeq analyzes a data set normalized within it for differential expression.Is there a way to get DESeq to analyze it for differential expression (DE)? Or if not (as I read that it needs raw read data and normalized data set cant be used) how can I use similar test of variance i.e. if I am not getting it wrong DESeq uses a chi square test can I replicate the same conditions(?) it uses elswhere to simulate a similar DE analysis as by DESeq?
How were the normalized numbers generated? You absolutely need raw counts to use DESeq2/edgeR/etc..
The reads were normalized using the KDMM normalization in JMPgenomics. I was actually looking for comparing my results with the DESeq statistical test for this normalized data.
Why not just use the raw counts fed into the KDMM algorithm? That would make more sense.
If you already have normalized values for your data, you can use limma to do a differential gene expression test. For instance, when I have raw counts, instead of using the
nbinomTest()function in DESeq to test for differential expression, I use variance stabilizing transformation (getVarianceStabilizedData()available in the DESeq package) to get normalized counts and then use limma to do a differential gene expression. So you can directly use limma on your normalized values.It'd be interesting to compare the results of that versus putting the normalized values through voom first. Since the variance stabilized values produced by DESeq2 aren't intended for statistical significance tests, I would be hesitant to just plug them into limma.
Oh I did not know that. I referred to this paper (a long time back) and according to it, vst+limma & voom+limma can be used for differential tests and perform similarly. Since then, I have been using vst+limma for such tests.
I'd forgotten that paper. I'd have to compare the underlying methods to see what else voom is doing that vst isn't.