edgeR and MA-plot question

[moem]

Hi,

I am running edgeR and used plotSmear to visualize my results. When coloring only the genes that are DE by FDR<0.05 I see that many genes have a higher fold change than some of the DE genes and that some of the DE genes have a low fold change. This is only the case when I use estimateGLMTagwiseDisp and not when I use estimateGLMCommonDisp. Does this mean that using common dispersion is the best? I only have cases and controls in my data so I guess it should be good enough but I always thought estimating tagwise dispersion was better?

Also:
It also seems like ~90 % of my genes are up-regulated. I tried all the different normalization options but the results seem to be the same. Can this be true or could something be wrong with my data? Any way of testing that?

[N311V]

I can't comment on your edgeR problem but maybe try using limma and see if a different approach gives the same result.

Regarding 90% up-regulated, what are the differences between your cases and controls? For example, were they sequenced at the same time, extracted at the same time, etc.

[mastal]

Does it make biological sense?

Is the condition you are comparing between cases and controls one where you might expect the majority of the genes to be upregulated?

[moem]

Thank you for your answers. I have run limma and DESeq now and it gives the same sort of answers with regards to the amount of up/down-regulation.

I was wondering if anyone could see if my code makes sense:

condition <- relevel(factor(input$condition),ref=”Control”)
design <- model.matrix(~condition)

“design” then lists “Intercept” and “ConditionCase”.

e <- DGEList(counts=countstable)
e <- calcNormFactors(e, method=”TMM”)
e <- estimateGLMCommonDisp(e, design)
e <- estimateGLMTrendedDisp(e, design)
e <- estimateGLMTagwiseDisp(e, design)
efit <- glmFit(e, design)
efit <- glmLRT(efit, coef="ConditionCase")

The logFCs are then mainly positive. My interpretation is then up-regulation in ConditionCase.

[N311V]

The result seems reasonable.

Is there any other reason you can think of that might cause systematic up regulation?

Were controls and treatments extracted at the same time? Sequenced at the same time? Are there age differences between treatments? Sex differences between treatments?

[moem]

Thanks for looking at my code.

Can't think of any reasons, but this is not a condition I know much about.

Extraction and sequencing at the same time and same sex. No systematic difference in age between groups.