EdgeR_normalization question

[xiansan]

I have two samples with very different composition: one sample contains many genes expressed at very high levels, which are not present in the other sample.

I tried the TMM method (EdgeR_TMM). The scaling factor is around 5. I am very new at this, so I am not sure whether it is still valid in this range. Helps are greatly appreciated!

[Wolfgang Huber]

Dear Xiansan

I think the answer of whether this is "valid" depends on your subsequent question. What is the aim of the experiment?

Best wishes
Wolfgang

[xiansan]

Dear Wolfgang,

Thanks so much for your help. Our data set is a time series, the aim is to detect distinct expression profiles that may correlate with distinct developmental processes.

Our first 5 samples have similar transcriptom composition (scaling factor = 1 to 2). The last 3 samples have different composition due to the activation of a storage program. TMM normalization indicated that reads in sample 8 should be multiplied by 5 (SF=5 between sample 1 and 8). I am more of a biologist, so I don't feel very comfortable deciding whether to use TMM to normalize samples with dramatically different compositions.

Thanks very much!

xiansan

[Wolfgang Huber]

Dear Xiansan

thanks. My approach here would be to find control genes, for which you know from other sources that they are not changing between the first 5 and the latter 3 samples; or, for which know by how much they are changing. And then determine the size factors based on these.

On a more philosophical note, in my experience the usefulness of differential expression analysis tends to be inversely related to the number of genes that are truly differentially expressed.

Best wishes
Wolfgang

[xiansan]

Dear Wolfgang,

Thanks very much for your suggestions, as well as your thoughts on our case. I suppose that we need to be more stringent when selecting differentially expressed genes in this particular case.

xiansan

[A Oshlack]

Hi Xiansan,

I suggest you visualise your data with something like an MAplot and see if that normalization factor looks about right. If you can see that there are really a group of genes that are very highly expressed in just one sample (group) then you can get factors that high and I have seen it before.

Cheers,
Alicia

[xiansan]

Hi Alicia,

Thanks for your comments. The MA plot did indicate that there are a number of genes expressed in one of the samples at very high levels. It is very reassuring to know that someone else had similar experiences. Thanks very much.

xiansan